YouTube playlist

• [Link] Sourcery with Molly O’Shea: NVIDIA GPU’s in Orbit: Inside Starcloud’s $2B Factory.

• [Link] Mach33 Financial Group: SPCX Q3 BEAT, 13 GW, DUAL CATCH | ep25

• [Link] Raoul Pal: One Code: How the World Actually Works. Exponential Age - intelligence per unit of energy. Love it. I like macroeconomics, but have a hard time (fully) grasping it. I’ve watched this video 3 times now, and plan to make it a staple in my AI learning path.

• [Link] My other favorite macroeconomist: Jordi Visser. The New Market Structure: AI Concentration and the Largest Hedge Fund Degrossing in 10 Years

• [Link] Next Big Future: AI Boom or AI Bust? SpaceX Could Make 20GW of AI in 2027. 10X of SpaceX Q3 Compute

  • Follow-up: Elon replied: Inside SpaceX’s 1.19 million GPU buildout

• [Link] TBPN - Gavin Baker on Anthropic’s S-1, GPU Prices, and SpaceX

The Capital Wall

Last spring on stage Eric Schmidt said “I think the real limit to AI is not energy. It’s actually cash. Because when you add up the cost of these, if you take 50, just round numbers, 50 billion a gigawatt, so 10 gigawatts is half a trillion dollars. How many companies, countries, and so forth can hand an industry a trillion dollars of capital? Very, very few.”

• [Link] Bank of America, Blackstone, Citi, and Morgan Stanley have announced a landmark $63 billion financing structure to support the expansion of AI compute capacity. Executed through a special purpose vehicle, the funding enables AI startup Anthropic to lease Broadcom’s advanced XPU AI accelerators and networking hardware starting in 2027.

• [Link] The next $1 trillion AI infrastructure opportunity. In her stock and industry analysis for the I/O Fund, lead technology analyst Beth Kindig highlights a major shift in AI spending beyond Big Tech hyperscalers, detailing five core findings: expanding $1T TAM beyond Hyperscalers, higher revenue per deployment, emergence of rack-Scale CPUs, massive CPU revenue opportunities, and Surging Liquid Cooling Adoption.

• [Link] Nvidia, AMD Commit $4B to Meta AI Buildout. Meta disclosed plans to raise $97.5 billion to support its continued investment in AI infrastructure; Nvidia announced a $3 billion investment, and AMD followed with a $1 billion commitment.

• [Link]Alphabet, Meta, Microsoft, Amazon and Oracle have committed more than $3 trillion to future AI spending on data centers, leases, equipment and energy, up from about $2.4 trillion for four of them in July.

• [Link] Startup funding deals are always worth a look as well: advanced silicon battery materials company Group 14 Technologies closed on $169 million, General Compute closed a $100 million deal, Foundry Robotics closed $100 million, and Bedrock Robotics closed on $600 million.

Data Centers

• [Link] Gun Range and Spam: SpaceX’s Two Massive New AI Data Centers. Both facilities are part of the rapidly expanding Colossus II supercomputing complex, which stretches across the Tennessee–Mississippi border. The two sites are expected to be operational later this year. Also check out this MeasuredAI Substack for analysis and some great pictures.

• [Link] Brian Wang argues that “dark GPUs,” AI chips already bought but stuck in unfinished data centers waiting for power, are the AI-era version of the dot-com era’s dark fiber. Using Elon’s estimate of up to 15 GW of stranded compute and Morgan Stanley’s projected US power shortfall of about 33 GW, he puts the problem at $300–500 billion. He says SpaceX is positioned to solve it: it builds powered data center shells in about 5 months for a fraction of the industry’s cost, and it could lease spare shells to customers who bring their own chips, which by his own modeling could be worth about $900 billion to SpaceX in 2027–28.

• [Link] Corning announced third-quarter 2026 results, with optical communications up 44% to $2.38 billion; Enterprise Networks grew sales 84%, and Gen AI product sales growing more than 100% year over year. In talking about the scale-out opportunity, CEO Wendell Weeks said, “Connecting those GPUs out down the rows to every other GPU that sits in the cluster. It is about 16 fibers per GPU in today’s architectures. 100% of the current scale of architectures are copper. If that was 100% optical, with scale-up, you increase to a 160 fibers per GPU.”

• [Link] Foxconn hopes to own 50% of the global AI server market within a couple years

• [Link] Microsoft acquired 616 acres of Dallas County (Iowa) farmland for $95.5 million (roughly $155,000 per acre) to develop a new data center campus. This is the same story that I reported earlier in the month about Woodward, Iowa, and the residents pushing back on the plans.

Quantum Computing

• [Link] Alice & Bob estimates that its error-resistant “cat qubit” design could break RSA-2048 internet encryption with about 22,600 physical qubits running for 3.14 days. That’s 40 times fewer qubits than Google’s 2025 estimate, and it says about 29,600 qubits could crack a Bitcoin-style elliptic-curve key in roughly an hour.

• [Link] IQM Quantum Computers fully entangled all 54 qubits on its cloud-accessible IQM Emerald processor at the same time, staying above the threshold that separates real entanglement from noise. Most studies use only 10–20 qubits, and Emerald users could previously count on about 30.

• [Link] Intel acquires Israeli quantum computing software company Classiq for about $2.3 billion. The Next Platform: Intel’s $2.3 billion bet on a quantum future is a decade in the making.

General Compute is building an “inference cloud for ultra-fast tokens,” a business model that turns the financing of non-NVIDIA AI chips into something customers can simply rent. In September, it signed a multi-year deal to deploy Cerebras wafer-scale systems for agentic coding, its largest hardware commitment and the first draw on its $400M Upper90 debt facility. The pitch is that alternative accelerators lose not on performance but because nobody finances and deploys them at scale. In October, it partnered with Positron AI to buy Atlas systems and committed to its next-generation Titan systems, built on Positron’s memory-heavy Asimov chip, so GPUs can process prompts while specialized chips generate tokens. On Oct 27, it announced a $100M Series A co-led by Blackbird and Abstract to fund that model, with first deployments going live in Q1 2027 and more than 50 MW of capacity in the pipeline.

[Link] PsiQuantum unveiled its Optical Circuit Switching (OCS) technology, demonstrating how the company’s advanced, low-loss silicon photonics platform can be applied to some of the most significant scaling challenges facing AI infrastructure. Their first OCS is an 8×8-radix demonstration vehicle built on the low-loss, silicon photonics platform that the company has spent nearly a decade developing for utility-scale quantum computing.

[Link] BlackSky signed a seven-figure annual extension with HEO to provide non-Earth imaging, meaning its satellites photograph other objects in orbit, for space domain awareness work across defense, intelligence, and commercial customers.

[Link] Xcel Energy has increased its five-year capex plan by 17% to $70B. On the grid buildout agenda: adding 9.2 GW of renewables, 2.3 GW of storage, and 3.3 GW of natural gas. The company also announced plans to build the upper Midwest’s largest battery storage portfolio - with the arrival of battery units at its Blue Lake site. Together with two planned projects in Becker, Minnesota, Xcel Energy expects to operate more than 735 megawatts of battery energy storage in Minnesota by the end of 2027.

[Link] California’s independent system operator, CAISO (California Independent System Operator), has proposed new rules for connecting large loads, such as data centers, to the state’s grid. Earlier in the week, the ISO said that its board approved a new proposal that seeks to enable large loads to interconnect faster to the transmission system.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

YouTube Playlist

• [Link] SemiAnalysis Ep. 039 - [Response] Our Numbers are “insane”. Here’s the proof (data centers). …and the Substack Datacenters 2027: The year of Acceleration, +50GW in a single year.

• [Link] The a16z show - Marc Andreessen: The Biggest Gap I’ve Seen in My Career. Erik Torenberg interviews Chris Dixon, Marc Andreessen, and Ben Horowitz about technology and expanding a16z crypto into a16z Frontier.

• [Link] The a16z show - Amazon CEO Andy Jassy on Reinventing Amazon for the AI Era. AWS Compute - 2026 compute is sold out, 2027 compute is largely sold out, and a good chunk of 2028 compute is sold out.

• [Link] Anthony Pompliano: Will Bitcoin Survive Quantum Computing? | Jordi Visser. I really enjoy every one of these weekly chats.

• [Link] Sundar Pichai on what’s next for Google Gemini, Waymo, and data centers in space.

• [Link] Next Big Future: AI Boom or AI Bust? SpaceX Could Make 20GW of AI in 2027. 10X of SpaceX Q3 Compute

• [Link] Undecided with Matt Farrell. How Iron-Air batteries work (and why they matter)

Data Centers

• [Link] A Common Sense Institute study estimates that data center owners have invested or announced at least $53 billion in Iowa through 2031: $15.1 billion so far and $38 billion in announced projects that would more than triple capacity, from 2,324 MW to about 7,581 MW. Ongoing operations matter more than construction, since existing sites already support about 12,647 jobs in 2026 and announced projects would support about 29,597 by 2031, with pay in the industry averaging $125,759, nearly double Iowa’s overall average. The report also notes that operating and planned data centers could draw nearly half of Iowa’s 2024 average electricity load, which is why Alliant and MidAmerican are adding new gas plants.

• [Link] At NVIDIA GTC in Berlin, Armada announced it will build factory-assembled, containerized “superintelligence factories” based on NVIDIA’s DSX POD reference design for Vera Rubin. The compute, networking, power, and liquid cooling are integrated and tested before shipping, so a site can be producing tokens in months instead of the years a conventional build takes.

• [Link] Crusoe introduced Spark Liquid, the sixth generation of its prefabricated modular AI data center. Each unit holds up to 3.5 MW of NVIDIA GB300 or Vera Rubin NVL72 racks, which is five times the compute density of earlier generations, and it uses closed-loop liquid cooling that doesn’t consume water. The units are built at Crusoe’s factory in Brighton, Colorado, with production starting in 2027, and can be deployed in months to add Crusoe Cloud capacity near users for low-latency inference.

• [Link] Crusoe announced a strategic partnership with Paraguayan energy and infrastructure developer Araico to build high-performance AI compute infrastructure powered by the Itaipú Hydroelectric Dam. The US government formalized its backing for an AI infrastructure partnership between US-based Crusoe and Paraguay’s Araico, the Paraguayan developer behind a planned US$12 billion AI data center complex.

• [Link] TikTok’s US joint venture is looking for multiple gigawatts of compute, possibly 4 GW by the early 2030s according to one source, as it retrains its algorithm on US data and builds AI products, though it says Oracle will stay its primary cloud provider.

• [Link] Microsoft files for a data center project in Floyd County, Georgia. Dubbed ATL80, all three buildings will total up to 1,659,000 sq ft of floor space, and the project will be valued at $360 million.

• [Link] Inside Prometheus: Meta’s Journey to a Gigawatt-Scale Supercluster. Meta finished the first phase of Prometheus, its gigawatt-scale AI supercluster in New Albany, Ohio, and used it to train its Muse models. It already links hundreds of thousands of NVIDIA GB200 and GB300 GPUs across several buildings and is heading toward about 1 million GPUs, with Vera Rubin racks coming next. On the networking side, there’s a dedicated Ethernet back-end fabric of up to 20,736 GPUs per building, tied together by an aggregation network spanning 2 to over 100 km that uses 800G-ZR coherent optics on the longer links.

• [Link] The Coolest Gear on the OCP 2026 Expo Floor - Data Center Richness. Awesome visual review from the expo floor at the Open Compute Project Global Summit.

• [Link] Brookfield to Invest Tens of Billions in Germany Over the Next Five Years. Brookfield’s investment activities will focus on energy, digital infrastructure, and industrial business sectors.

• [Link] Nebius got a $3.5 billion, 364-day senior secured loan from Citi to pay for data centers and GPUs, priced at Term SOFR plus 2.75% and rising to as much as 3.50%. It works as a short-term bridge alongside July’s roughly $775 million GPU-backed loan and August’s $5.75 billion convertible notes.

• [Link] DayOne, a SoftBank-backed data center operator based in Singapore, filed on Oct 5 for a Nasdaq IPO under the ticker DODC and may raise up to $5 billion, with Morgan Stanley, J.P. Morgan, BofA and Citi as underwriters.

Energy

• [Link] Google and Voltus expanded their deal tenfold, from 100 MW to a 1 GW virtual power plant across the PJM grid region, where Voltus pools smart thermostats, batteries and industrial equipment at homes and businesses that cut their power use during peak demand. Google funds it, participants get bill savings and cash payments, and Google says it’s one of the largest distributed capacity portfolios any hyperscaler has deployed.

• [Link] Elisa Wood points out that the Google/Voltus deal wasn’t even the biggest virtual power plant in the week, as two days earlier Verantum and Yotta Energy announced plans to assemble a 5-GW virtual power plant from commercial buildings across North America.

• [Link] Blackstone closed its Energy Transition Partners V fund at its $8.7 billion hard cap, more than 50% larger than the previous fund, to back power generation, transmission and energy equipment businesses meeting growing electricity demand. It has already funded three platforms (Kindle, Fisterra and Viridon) and made three acquisitions, including Flow Control Holdings, which makes liquid-cooling components for data centers.

• [Link] The Department of Energy announced about $967 million in new funding through its Grid Resilience State and Tribal Formula Grant program to help states, territories, and tribes upgrade grid infrastructure. The money goes to locally chosen projects that prevent outages, improve reliability, and keep up with growing electricity demand, and DOE frames it as a way to lower power costs for families and businesses.

• [Link] Commonwealth Fusion Systems signed a letter of intent with the Commerce Department for a proposed $85 million CHIPS Act R&D award to develop semiconductors and power electronics for its ARC fusion power plants. The work covers chips for robotic maintenance systems, better radio-frequency power generators and fast embedded controllers to manage fusion plasma at power-plant scale, and CFS expects it to draw more suppliers into the fusion industry.

• [Link] Helion, a Washington-based fusion energy company, announced it has signed a letter of intent with the U.S. Department of Commerce to receive up to $175 million in funding through the CHIPS Research and Development Office to conduct R&D critical for the development and manufacturing of next-generation semiconductor technologies for commercial fusion energy.

[Link] Beyond transistor limits: Hybrid magnetic-sound waves could boost computing speed and efficiency. University of Michigan and University of Wisconsin researchers, with up to $6 million from DARPA, are designing a transistor alternative called M3BOLT that they say could make computers 100 to 1,000 times faster and more energy efficient. The reason is that today’s transistors are expected to hit their efficiency limit around 2037: chips get better by shrinking parts, but at a few nanometers electrons start leaking through even when a transistor is off, which wastes power.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

YouTube playlist

• [Link] ServeTheHome: Exclusive INSIDE the Secretive AWS Networking Lab

• [Link] All-In Podcast: CoreWeave’s debt selloff. Chamath does a deep dive into whether the AI buildout is creating a private equity bubble.

  • I missed this a few weeks back, but Silicon Angle has a thorough analysis of CoreWeave.

• [Link] Anastasi In Tech explores the Chinese AI infrastructure boom and the Huawei Hi-ONE, a new optical interconnect engine designed for the next generation of SuperPods.

Open Compute Project Global Summit

• [Link] Data Center Richness: Agentic AI is Hungry for Hardware. AMD, NVIDIA, and Meta showcase innovations at the OCP Summit. AMD’s Dr. Lisa Su delivered the keynote and launched the company’s new Helion rack.

• [Link] The OCP Foundation announced a sharp increase in IDC’s forecast for worldwide spending on OCP-recognized IT equipment. The 2026 IDC Market Impact Study projects annual spending of approximately $800 billion in 2029, up from the $295 billion forecast for the same year in last year’s study.

• [Link] HPCwire Infrastructure highlights from OCP Global Summit.

• [Link] Astera Labs unveiled Orion, a modular rack design for AI inference that lets neoclouds and enterprises use whichever GPUs and CPUs they choose, connected by a PCIe switch fabric that scales from 8 to 160 accelerators per pod, with racks sold through OEM and ODM partners starting in Q1 2027.

Iowa

• [Link] Turned down in Ames, Iowa, LightEdge has looked just west of Ames and submitted a data center site plan to the City of Boone, where Alliant Energy says up to 46 MW could eventually be available.

• [Link] Microsoft bought more than 600 acres of industrial-zoned land in Woodward, Iowa, for a future data center, and it unveiled the plan at a tense residents-only meeting where people objected to having no vote. Design alone is expected to take about two years before construction starts. Woodward is roughly 20 miles north of the data center campuses Microsoft has in Des Moines.

Data Centers

• [Link] NTT Data Group is in final talks to build Anthropic’s first dedicated data center in Japan, according to Nikkei’s sources. The Osaka facility could cost about $6.3 billion, would be one of the largest in the country by processing capacity, and could partly open by 2028.

• [Link] Microsoft is scaling its in-house Azure Maia AI chips to multi-gigawatt deployments starting in 2027, making Maia the main system that runs inference for its products. The next-generation Maia 300 arrives in 2027 and promises 40% more tokens per megawatt than leading GPU-based systems, with 96 accelerators in a 167 kW rack linked over Ethernet-based networking.

• [Link] Voyant Photonics announced plans to develop a chip-scale optical circuit switching (OCS) engine for AI data center networks, adapting its production-proven silicon photonic switching architecture that has been industrialized for its FMCW LiDAR platform.

• [Link] A Sam.gov contract opportunity details U.S. DoD intent to develop a mission-critical command and control center and data center for the US Space Command. The $4 billion, 700,000 sq ft facility will be located on 64 acres at the Redstone Arsenal Army base in Alabama.

• [Link] Skeleton Technologies announced a strategic investment from Siemens, bringing Skeleton’s valuation to more than $1 billion. The investment will primarily accelerate the expansion of Skeleton’s US manufacturing footprint and R&D activities in Silicon Valley to meet growing demand from hyperscalers for AI data-center power systems.

• [Link] Senators Alex Padilla and Eric Schmitt introduced the bipartisan Combating Critical Infrastructure Vandalism Act, which would make willful theft or damage of copper, fiber, and coax lines used for telecom and information services a federal crime, including on privately owned broadband networks.

• [Link] Seven Senate Finance Committee Democrats, led by Ron Wyden, want Meta to explain by Oct 31 how it cut its federal taxes by over $6.6 billion in three years with research credits, which they say appear tied largely to its AI data center build-out, raising the question of whether data center construction counts as research.

• [Link] Engineering at Meta: Inside Petal: Building the World’s First Petabit-Class Transoceanic Subsea Cable. Expected to be in service in 2029 Petal will be the first subsea cable to deliver petabit capacity at transoceanic distances, connecting France and the United States over approximately 7,000 km (4,300 miles).

Quantum

• [Link] Anthropic announced that Claude designed more resource-efficient quantum circuits in collaboration with researchers from Harvard and QuEra Computing. This is software research aimed at reducing what future quantum computers need, and not an announcement that Anthropic has built a quantum processor. This work was led by Andi Gu as part of Anthropic’s STEM Fellows Program, with mentorship from Alek Kemeny, a quantum physicist at Anthropic, in collaboration with Susanne Yelin (Harvard University) and Mikhail Lukin (Harvard University and QuEra Computing). Read the full paper on arXiv. … or ask Claude to translate the research into something a little easier to grasp.

• [Link] Google Says it cuts quantum error rates by 40x with ‘magic state cultivation’. Google Quantum AI used a technique called magic state cultivation on its Willow processor to cut errors 40x while producing the special “magic states” quantum computers need for complex operations, reaching a record 99.99% fidelity, according to results published in Science.

• [Link] IonQ agreed to buy Zero Trust security company Elisity in an all-stock deal worth up to $1.3 billion, pairing its post-quantum cryptography and quantum key distribution with Elisity’s microsegmentation platform to protect enterprise and government networks against current and quantum-era threats. IonQ has also secured $2 billion through an equity offering managed by Heights Capital Management.

• [Link] Cisco‘s Outshift research unit unveiled a prototype Quantum Network Controller, a control layer that automates the setup of entangled connections between quantum nodes, which is still done by hand today, and offers it to applications as “Entanglement-as-a-Service” through an API that works across vendors and qubit types. It also showed a network-aware quantum compiler that splits large circuits across multiple processors and works with the controller to deliver entanglement when it’s needed.

Energy

• [Link] AEP plans to invest about $4 billion in a 702 MW Bloom Energy fuel cell plant near Cheyenne, Wyoming, that will supply Google under a 20-year power purchase agreement, with construction starting in 2026 and full operation expected by the end of 2028.

• [Link] The Department of Energy announced two grid-expansion efforts this week focused on getting more power out of existing infrastructure. On Oct 13th the DOE said it will support the Tennessee Valley Authority on 84 transmission and distribution projects across eight states, a roughly $581 million effort that includes $250 million in federal funding, adds about 2.5 GW of capacity, and supports around 800 jobs. On Oct 16th DOE said it intends to give up to $50 million in Defense Production Act funding to a Virginia and West Virginia project that uses reconductoring and real-time line monitoring to add about 290 MW to Appalachian Power’s Virginia system, one to four years faster than conventional upgrades.

• [Link] LNG exporter Venture Global signed frame agreements with GE Vernova and Baker Hughes for up to 16 new 720 MW gas-fired “power island” units, expanding its off-grid, behind-the-meter power network beyond the 4.5 GW it already uses for its own facilities. The company says the expansion could support up to 10 GW of new data center compute on land it owns or controls by the end of 2034, without drawing on the public grid, as part of its pitch to become a vertically integrated “AI utility.”

• [Link] The White House Office of Science and Technology Policy (OSTP) hosted the “Science: A New Golden Age Summit” Thursday. The biggest space news came from NASA Administrator Jared Isaacman and Energy Secretary Chris Wright, who signed an agreement to accelerate cooperation between the two agencies on space nuclear power and propulsion. Nuclear power plays a central role in the agency’s plans for its Moon Base, on which Isaacman also provided an update.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

Iowa

[Link] In advance of a public meeting later this month, Google has mailed a letter to Palo, Iowa residents with some details of the 545-acre data center proposal. “We have been exploring the possibility of a data center in Palo, and I wanted to help share more information so the community can hear directly from us about what this project will look like and how it can benefit Palo,” Steve Walter, Google’s vice president of global infrastructure and energy, wrote in the letter. Google announced plans in May 2025 to invest $7 billion in its Cedar Rapids and Council Bluffs Iowa data centers, with a proposed Palo campus representing further investment beyond that initial amount. Google also launched PaloDataCenter.com to provide information and updates on the proposal.

I spent some time driving around Cedar Rapids today to get a firsthand look at the QTS and Google data center sites. The scale of development is really impressive. Both projects are well underway, buzzing with activity, and backed by a huge footprint of equipment and construction crews.

Data Centers

• [Link] Google enters 3.6GW deal with Constellation Energy. The deal includes funding for 890 megawatts of nuclear reactor upgrades to fuel its expanding AI data center operations. In exchange, Constellation will utilize Google Cloud and Gemini AI technologies to support and optimize its power grid operations.

• [Link] AWS has filed a Preliminary Land Development Plan (PLDP) for a 36-building data center campus in Homer City, Pennsylvania. The site will be powered by a 4.5GW natural gas power generation facility being constructed by Homer City Redevelopment.

• [Link] CoreWeave has expanded into India by partnering with AdaniConneX to deploy a 240-megawatt AI data center capacity equipped with NVIDIA Vera Rubin infrastructure at its campus in Navi Mumbai.

• [Link] French AI infrastructure company Sesterce has announced plans for a €10 billion, renewable-powered AI supercomputing campus at the former Kaipola paper mill site in Jämsä, Finland, starting with a 200 MW first phase set to begin construction in 2026.

• [Link] Oxide Computer Company has raised $445 million in Series D funding, led by Eclipse and featuring new investment from AMD Ventures, following its achievement of profitability earlier this year. The company will use the new capital to scale its manufacturing capacity to meet overwhelming enterprise demand for its on-premises, full-stack cloud computing systems.

• [Link] Amazon is reportedly in advanced discussions to acquire Israeli AI startup Decart in a deal potentially worth up to $7 billion, which could close in the coming days. Decart’s DOS software, which optimizes neural networks to easily shift AI workloads across different chips, could significantly benefit Amazon Web Services and its custom Trainium processors.

Space

• [Link] Very big Starmind news, and the networking part is what gets me. On X @aaronburnett said:

  This is the first we've seen SpaceX discuss formation flying and have confirmed they are pursuing a hybrid model of both larger sats (250kw) and 10 sats flying in close formation. Our original math suggested this needed to be in sub 750m proximity. historically starlink proximity has been measured in kms.
  
  Formation flying of 10 sats suggests coherence up to 720 GPUs (which is massive; I think the largest tested deepseek cluster was ~300). if architected to maximize token density it could handle very large models with maximum intelligence density.

  That’s the interconnect problem every data center fights, now in orbit. 🛰️ See this Mach 33 article for more.

• [Link] SpaceX announced an agreement to acquire a nationwide low-band spectrum license portfolio of up to 14 megahertz of paired spectrum in the 800 MHz band. This prime low-band spectrum addresses one of the key remaining technical gaps that will pave the way for Starlink Mobile to become a major mobile carrier in the US. The Federal Communications Commission also approved Starlink Mobile’s Gen2 constellation application this week, authorizing SpaceX to launch 15,000 satellites that are optimized for 2 GHz spectrum globally. [Disclosure: I own 1 share of SpaceX… but would like to own a LOT more]

• [Link] AST SpaceMobile Podcast: SpaceX buys Grain Management spectrum explained. @spacanpanman reacts to the news that SpaceX is acquiring Grain Management’s low-band spectrum, alongside a major FCC order released the same day. He explains why the deal looks negative for AT&T, Verizon, and T-Mobile at first glance, but strengthens the strategic case for AST SpaceMobile.

• [Link] Axiom Space has advanced its orbital compute strategy by deploying two Axiom Resilient Compute (ARC) nodes aboard Kepler Communications‘ optical data relay constellation. Designed for edge AI processing and post-quantum cryptography, the ARC platform reduces reliance on ground bandwidth while securing space-based communications against future quantum decryption threats.

[Link] IonQ has demonstrated a photonic quantum interconnect achieving over 1,000 entanglement connections per second between a trapped ion qubit and a solid-state silicon vacancy quantum memory. This breakthrough addresses a key networking bottleneck to enable scalable, distributed quantum computing and networked quantum data centers.

[Link] PsiQuantum announced that the company has received a conditional commitment for a loan of up to $150 million from the U.S. Department of War’s Office of Strategic Capital (OSC), which would expand the company’s advanced manufacturing and prototyping capabilities in the United States.

[Link] The Linux Foundation has launched OpenGrid, an open-source initiative designed to create a shared data and software foundation to modernize power grid planning worldwide. By establishing open standards, interoperable APIs, and model-ready datasets, OpenGrid aims to help utilities and regulators accelerate grid infrastructure expansion to keep pace with growing global energy demands.

[Link] Arista Networks announced an open, liquid-cooled Ethernet rack-scale portfolio for scale-up and scale-out AI infrastructure developed alongside industry partners like AMD, Arm, Broadcom, Meta, and Microsoft. The new architectures support up to 144 accelerators per rack using standards-based Ethernet to provide cloud providers with a flexible alternative to proprietary compute systems.

[Link] VEIR raised $110 million in an oversubscribed Series C financing round co-led by Matter Venture Partners and Tyche Partners, bringing its total funding to $225 million. The funds will be used to accelerate the commercialization of the company’s superconducting power delivery systems designed for AI data centers. VEIR supplies an integrated power-delivery system rather than just a superconducting cable. Its platform combines cables with cryogenic cooling, controls, termination units, and monitoring.

[Link] The U.S. Department of Energy announced $159 million across 12 Phase II Genesis Mission awards to apply artificial intelligence and advanced computing to complex scientific fields, including microelectronics, quantum computing, particle physics, and biotechnology. Alongside the research grants, the DOE launched new educational fellowship programs, including a Genesis Mission Graduate Fellowship for 4-year PhD tracks and a high school fellowship - to train the next generation of American researchers using national laboratory and industry experiences.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

Data Centers

• [Link] AMD announced that it has entered into a definitive agreement to acquire World Labs, an AI model and research lab led by AI pioneer Dr. Fei-Fei Li. The acquisition will bring a world-class team of researchers and model experts to AMD, strengthening its ability to develop AI hardware, software, and systems around the needs of emerging models and applications. While large language models master the patterns of text to simulate communication, world models represent the next evolutionary leap by building an internal simulation of physical reality. This shift moves AI from merely predicting the next plausible word to understanding spatial dynamics, cause-and-effect, and the underlying rules of how the real world actually functions. An LLM was used, but not harmed, in the formation of those last two sentences.

• [Link] Crusoe announced it is developing a data center campus in Armstrong County, Texas, to power Google’s AI infrastructure directly using adjacent wind energy. The water-efficient facility began construction in June 2025 and utilizes a closed-loop liquid cooling system to minimize its local water footprint. The campus has a direct connection to Serena’s adjacent Goodnight 1 (265.5 MW) wind farm, which is operational, and Goodnight 2 (265.5 MW) wind farm, which is currently under construction.

• [Link] Crusoe files for $4.8B data center campus in Jayton, Texas. Construction will begin at the end of January 2027, with the buildings expected to go live in May and July 2029.

• [Link] Lambda announced plans to build a new AI data center in Mayes County, Oklahoma, at the MidAmerica Industrial Park. The project is expected to generate $500 million in local tax revenue over the next decade, create up to 1,100 jobs, and utilize closed-loop cooling to conserve water while paying 100% of its energy costs under state ratepayer protection laws.

• [Link] Palantir and Armada have partnered to deploy sovereign AI infrastructure by integrating Palantir’s Sovereign AI Operating System with Armada’s U.S.-manufactured Galleon modular data centers. This joint solution enables enterprise and government clients to run and adapt AI models on physically owned, fully controlled, and resilient compute infrastructure, including air-gapped environments, in months rather than years.

• [Link] Hewlett Packard Enterprise has secured a $1.2 billion order from cloud infrastructure company Vultr to deploy its first AMD Helios AI Rack systems across U.S. data centers. The integrated, liquid-cooled infrastructure combines AMD GPUs with purpose-built HPE Juniper Networking Ethernet switches to power demanding enterprise AI training and inference workloads.

• [Link] Forbes: There’s no AI bubble. Billionaire Bob Clark, founder of construction giant Clayco, refutes concerns of an artificial intelligence bubble, arguing that massive data centers and advanced computing facilities will continue to be built regardless of market skepticism. Driven by the surging demand for AI and next-generation tech infrastructure, Clark’s real estate and construction ventures - including major projects like Chicago’s $12 billion Quantum Shore campus - have propelled his net worth to quadruple over three years.

• [Link] Data Center Alley: Lessons from the Front Lines. Rich Miller interview with Buddy Rizer (Loudoun Executive Director of Economic Development) on how Loudoun County, Virginia became the global capital of internet traffic.

• [Link] Virginia’s new energy plan prioritizes distributed energy resources - such as local solar, battery storage, and virtual power plants - to manage rising electricity demand and cut system costs through 2050. It also calls for data centers to adopt demand flexibility and help finance these customer-sited resources rather than shifting grid expansion costs onto households.

• [Link] Startup Valar Atomics has proposed “Project Beehive,” a massive development seeking thousands of acres of public and state land in central Utah to pair an AI data center with 456 small nuclear reactors capable of generating 9.6 gigawatts of power. While supporters view the project as a major economic and clean-energy boost, environmental groups and local residents have raised concerns regarding waste storage, water usage, and the use of public lands.

Space

🚀 There were six rocket launches last week, with SpaceX completing three launches in less than 13 hours!

• [Link] Mach33: Starship’s first orbital cargo flight cost an estimated $4,915 per kilogram. 26 Starlink V3 satellites were deployed. A V3 satellite carries 1 Tbps, about three times more capacity per kilogram than a V2 Mini Optimized satellite at 0.096 Tbps. Falcon 9’s $25M cost midpoint sits inside a $20M–$46M range, which spans $1,199–$2,759/kg and $7.2M–$16.5M per Tbps.

• [Link] Planet Labs announced the successful launch of 20 satellites: the first demo mission for Google’s Project Suncatcher (test customer TPUs); Tanager-2 (captures imagery in 426 contiguous bands with ~5 nm spectral resolution at 30 m spatial resolution), its latest hyperspectral satellite; and 18 SuperDoves.

• [Link] Star Catcher Industries launched its prototype power node, Protostar, aboard SpaceX’s Transporter-18 mission to execute the world’s first power-beaming demonstration between two untethered spacecraft in orbit. Protostar is the first end-to-end prototype of the company’s core technology stack on orbit: energy harvesting, satellite acquisition and tracking, and energy transmission.

• [Link] Sophia Space and Redwire have signed a Memorandum of Understanding to collaborate on scalable orbital data center infrastructure. The partnership combines Sophia Space’s modular computing architecture with Redwire’s power, thermal management, and space system integration expertise to develop next-generation in-orbit AI and edge-computing capabilities.

Energy

[Link] The United States and South Korea have entered into a strategic framework agreement committing up to $120 billion in Korean investment to support the construction of eight nuclear reactors using Westinghouse technology on U.S. federal sites.

[Link] Global energy storage company HiTHIUM unveiled its next-generation ∞Power N4.0MWh integrated sodium-ion energy storage system and ∞Cell N785Ah battery, which features a 30-year service life and is planned for full-scale production in 2027 to accelerate sodium-ion deployment.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

[Link] The trillion-dollar buildout lands in Las Vegas: Yotta 2026 opens this week

[Link] Akamai has signed a seven-year, $11.6 billion cloud infrastructure agreement with Anthropic to power the AI company’s expanding CPU workloads, with options to extend the deal by up to $9 billion. As part of the strategic partnership, Akamai granted Anthropic equity warrants for up to approximately 5% of its common stock, tied to initial execution and future spending milestones.

[Link] A consortium backed by BlackRock and IFM Investors is in exclusive talks to buy Stack Infrastructure’s Asia Pacific data center portfolio for ‌up to $25 billion, Bloomberg News reported, citing people familiar with the matter.

[Link] Applied Digital has revealed that its $3.2 billion Delta Forge 2 AI data center will be located near Brookwood, Alabama. The tenant is said to be an “investment grade hyperscaler.”

[Link] Switch has submitted plans for a roughly 175-acre data center campus in North Las Vegas, according to a letter of intent from the company obtained by KTNV Las Vegas.

[Link] Crusoe and Thinking Machines Lab have entered into a $65 million annual agreement to power production inference for open AI models using Crusoe Cloud. Thinking Machines Lab will run its workloads, including its Inkling models and GLM variants, on dedicated NVIDIA HGX B200 clusters optimized for high throughput and cost efficiency.

[Link] Nscale announced its $3.36 billion raise via convertible loan notes, led by Third Point. The convertible funding round was also supported by new and existing investors, including NVIDIA, funds managed by Apollo, Citadel, Hudson Bay Capital, Abu Dhabi Investment Council, and 8090 Industries.

[Link] The Chinese AI Infrastructure Boom: Introducing the SemiAnalysis China Datacenter Model. Tracking over 1,000 facilities across 60+ operators.

Energy

• [Link] The U.S. Department of Energy announced $5.25 billion in combined federal and recipient funding for 31 grid-improvement projects across 26 states under the new SPARK initiative. By upgrading and rebuilding existing transmission infrastructure rather than constructing new corridors, the investments aim to unlock over 23 gigawatts of additional grid capacity.

• [Link] Fervo Energy announced that Cape Station, its flagship geothermal development in Beaver County, Utah, has achieved First Power, marking the first time a utility-scale enhanced geothermal development has reached this milestone anywhere in the world. Cape Station Phase I is an approximately 100 MW installation comprising three 33-megawatt GeoBlocks (modular unit design for utility-scale Enhanced Geothermal Systems), the first of which has achieved First Power.

• [Link] Interconnection.fyi: Queue Update. Incredible report; loaded with data. 44 projects adding 9.45GW, and 33 projects withdraw 5.99 GW. Active capacity stands at 1.98 TW across 8,993 projects. The active national interconnection queue remains heavily dominated by five technologies: battery, solar, gas, solar-plus-battery hybrids, and wind, which collectively account for roughly 92% of all active capacity.

• [Link] Verse announced that enterprise energy buyers now manage 30 gigawatts of power on its Aria platform, tripling its capacity under management since early 2026. This growth is driven by demand from data center operators using Verse’s software and dispatchable assets to bypass grid connection delays and optimize complex energy portfolios.

• ⭐ I have run across so much data about the power markets lately that I stumbled into enabling MCP server connectivity to drive AI-powered searches. MCP (Model Context Protocol) servers are programs that expose specific capabilities to AI applications through standardized protocol interfaces. I’ve discovered two so far. Kardashev Labs (awesome name) offers live demand, queues, wholesale prices, and curtailment in one place. Sourced from public ISO and EIA feeds. Ask the Grid MCP can read live prices, load, generation, outages, plants, and pricing nodes across eight US power markets.

Space

• [Link] Behind Project Suncatcher. Google is launching a prototype satellite equipped with Tensor Processing Units (TPUs) to evaluate how its AI hardware handles orbital radiation, rocket launch vibrations, and vacuum cooling challenges. The blog post update covers the project overview, how AI hardware survives in space, cooling AI chips in space, and keeping satellites connected with lasers. Google’s next stated milestone is 2027, when it plans to put two satellites in orbit to test the precise, high-bandwidth laser links needed to form a distributed compute cluster. Sundar Pichai noted the first test in orbit, saying that it was built in partnership with Planet, and will be launching a prototype satellite aboard the Transporter-18 rideshare mission with SpaceX.

• [Link] Northrop Grumman has been selected by the Department of War Innovation Unit (DIU) to develop a commercially derived prototype for the Geosynchronous High-Resolution Optical Space-Based Tactical Reconnaissance (GHOST-R) mission. The program will provide the U.S. Space Force with a lower-cost spacecraft designed to augment existing space domain awareness capabilities in geosynchronous orbit.

• [Link] Intellian Technologies and Network Innovations have partnered under a multi-year strategic agreement to develop and distribute multi-band, multi-orbit Wideband Global SATCOM (WGS) tactical flyaway terminals designed to provide flexible, resilient satellite connectivity for military and government applications.

[Link] Quantum XChange has partnered with Carahsoft Technology, designating Carahsoft as its Master Government Aggregator. Through this partnership, Quantum XChange’s crypto-agile and FIPS-validated post-quantum security solutions will be made available to federal, state, and local government agencies to support their transition to quantum-safe cryptography.

[Link] IonQ demonstrated a quantum generative model—tested on its hardware using real satellite sensor data—that outperformed classical methods in detecting changes from complex, high-resolution radar imagery. The approach successfully handles difficult non-Gaussian pixel statistics, offering valuable analytics for applications like defense and disaster response.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

My weekly inspiration (videos)

• [Link] Ron Baron on Tesla FSD, SpaceX’s Starlink, and Data Centers in Space (CNBC). Some of what he said reminded me of what Jordi Visser has said (repeatedly) about agentic developments. “They (carriers) could not possibly have foreseen agentic demand coming for communication. They didn’t see it, so they haven’t been investing. All they want to do is keep existing customers, retire, and pay a dividend. They aren’t risking anything. [You think satellite wireless is eventually going to wipe out all the cable and fiber they’ve laid in the ground?] It costs $16,000 a mile to put wire in the ground, and when it breaks, you have no idea how to find it. ….

  and, In my opinion, the government is going to come around to these telcos and say, ‘Listen guys, you can’t just keep that spectrum when you’re not going to be able to use it and we need it.”

• [Link] Starship - The Holy Grail of Rocketry

• [Link] Bitcoin Is About To Go PARABOLIC Because Of AI Agents. Awesome conversation with Anthony Pompliano and Jordi Visser.

Data Centers

• [Link] Crusoe announced the initial closing of a $3.9 billion Series F funding round at a $30.9 billion post-money valuation. Crusoe has announced over 6 gigawatts of upcoming contracted data center capacity, representing more than $140 billion in revenue. This includes a Texas data center campus being developed specifically for Google, as reported by The Wall Street Journal.

• [Link] Emerald AI, Google, and NVIDIA have launched the AI Energy Management Alliance (AEMA) to advance data centers capable of dynamically adjusting electricity use based on grid conditions. This initiative aims to accelerate the growth of AI infrastructure while protecting grid reliability and energy affordability.

• [Link] Microsoft expands plans outside of Atlanta, Georgia, developing two three-story buildings totaling 910,000 sq ft in Union City.

• [Link] Vantage has closed a $2 billion financing facility to support early-stage development across its North American platform.

• [Link] ODATA, an Aligned Data Centers company, announced a $630 million first-phase deployment of DeltaFlow liquid cooling technology across its Brazil and Mexico campuses to sustainably support high-density, high-performance digital workloads in Latin America.

• [Link] Equinix Boosts Annual Data Center Investments to $7 Billion. Equinix plans to invest $5 billion to $7 billion annually to build out its data center footprint, aiming to double the total capacity delivered over its entire 27-year history within the next three to four years.

• [Link] Schneider Electric has expanded its AI-ready infrastructure portfolio by introducing standardized, prefabricated power modules and skids that deliver up to 2.5MW of scalable power capacity to rapidly support GPU-intensive data center workloads.

• [Link] Rune unveiled RELIC (Renewable Energy Linked Intelligent Compute), a modular compute system that installs directly at solar generation sites, turning wasted power into high-performance compute capacity. The company also closed a $40 million funding round led by Spark Capital.

• [Link] Nscale Limited filed S-1 with the U.S. Securities and Exchange Commission relating to a proposed initial public offering of its ordinary shares.

• [Link] Upscale AI launched with over $100 million in seed funding to build ultra-low-latency, high-performance AI networking infrastructure. Co-led by Mayfield and Maverick Silicon, the company designs open-standards-based silicon, systems, and software to connect specialized AI compute clusters and eliminate data center bottlenecks.

• A few SemiAnalysis videos/articles

  • [Link] Local Moratoriums: Where Is the Risk?

  • [Link] Ep. 032 - 300 Datacenter Bans, 6 Projects Delayed: Moratoriums Explained.

  • [Link] Ep. 029 - Modular Data Centers Cut Build Time to 12 Months

• [Link] Benjamin Grant shared photos with PetaPixel for before and after pictures of some large data center build-outs. xAI Colossus in Memphis, Open AI in Abilene, and Meta in Alabama.

Chips

• [Link] Edge AI semiconductor company Ambarella has announced the X7, its first standalone AI accelerator designed to bring low-power, edge AI performance to existing Arm and x86 host processors without requiring a full system redesign. Running on 2 to 5 watts, the PCIe and USB-compatible chip allows legacy embedded systems—such as security cameras, robotics, and industrial controllers - to add advanced “physical AI” and multi-model capabilities. The company also announced a strategic partnership with ZEDEDA to bring secure, cloud-orchestrated AI to the devices where the physical world meets AI: cameras, robots, vehicles, industrial systems, and edge appliances.

• [Link] Aeluma executed an agreement with Sumitomo Chemical Advanced Technologies (SCAT) to accelerate development and production of photonics wafers for the AI datacom market.

• [Link] Caltech’s tiny new chip can steer light in 74 quadrillionths of a second. Caltech researchers have created a nanoscale silicon metasurface that lets one beam of light redirect another in just 74 femtoseconds, using light itself rather than electronics to control the switch. The advance could point toward far faster, compact optical components for communications, computing, LiDAR, and imaging - though it remains an early laboratory demonstration.

• [Link] Inside its Hillsboro, Oregon labs, Intel researchers are developing novel photoresist materials, such as metal oxide resists, to support the ultra-thin, precise chemical layers required for High-NA EUV lithography. By advancing this chemistry, Intel aims to push semiconductor resolution down to 8 nanometers, unlocking the full transistor-scaling potential of next-generation EUV tools.

Quantum

• [Link] Researchers from IonQ, Oak Ridge National Laboratory, NVIDIA, and the University of Tennessee, Knoxville demonstrated a generative AI approach that directly writes quantum optimization circuits. By training a transformer model to replace the traditional trial-and-error parameter tuning loop, the team significantly improved solution quality while keeping runtime nearly constant as subproblem sizes grew.

• [Link] Anderon, an IBM company, has finalized a $1 billion CHIPS and Science Act award with the U.S. Department of Commerce to scale up quantum wafer research and domestic manufacturing capabilities. Operating out of Albany, New York, the 300-millimeter pure-play quantum foundry - supported by an additional $1 billion investment from IBM - has already begun running its first quantum wafers through the facility.

• [Link] Infleqtion has integrated its open-source qLDPC error-correction library with NVIDIA’s CUDA-Q Logical platform to advance fault-tolerant quantum computing software. Using an early-access version of the platform, researchers successfully validated a quantum error correction code that achieves a fivefold reduction in physical data qubit overhead compared to traditional surface-code approaches.

Energy

• ⭐ [Link] Ember Futures’ annual slidedeck: The Age of Power. Incredible. It sets out how the electric and information revolutions are converging, and traces what this means for the world. The world runs on two foundations: energy and information. Energy does work; information directs it. The digital revolution shifted information onto electrons; the electrotech revolution is shifting energy onto electrons as well. Bookmarking this site for future reference.

• [Link] Bloom Energy released a special report “The New Rules of AI Power,” showing that its unique 800V DC-native fuel cell solution can reduce non-compute capital expenditures for a 1 GW AI data center by $3.6 billion, or 27%, and five-year total cost of ownership by $5.5 billion, or 9%.

• [Link] Oracle announced a series of investments in Texas-based wind energy generation projects anticipated to deliver more than 1.7 GW of electricity, advancing the company’s goal to match 100% of the electricity use at its AI data centers with carbon-free energy.

• [Link] Utility-scale energy storage developer esVolta has upsized its corporate credit facility to $450 million in a transaction led by Nomura Securities International. The funding will be used for project development, equipment procurement, and pre-construction costs to expand its battery storage portfolio across key U.S. markets.

• [Link] Avanza Energy: The $900 Billion Question: How States Are Deciding Whether Data Centers Lower Your Power Bill or Raise It. Whether large-scale data centers lower or raise residential electricity bills depends entirely on utility tariff design and cost allocation. While correct incremental pricing and proper rate structures can turn new load into a rate-reducing asset, improper tariffs risk shifting massive grid infrastructure costs onto existing ratepayers.

• [Link] A 9.2 GWh demonstration by Quintrace, esVolta, and LevelTen Energy proves batteries can time-shift granular carbon-free energy certificates. A Google-backed pilot program successfully demonstrated a new commercial structure that allows battery storage operators to “time-shift” solar energy and its hourly-verified environmental certificates to periods of high electricity demand. This approach unlocks a novel environmental-attribute revenue stream for battery owners while supporting corporate goals for around-the-clock carbon-free energy.

• [Link] TransGrid Energy has launched EnergyFluo, an agentic AI energy management system designed to help data centers and large-load customers optimize power usage, reduce operational costs by 5% to 25%, and operate as flexible grid resources. By pairing physics-informed models with facility operating procedures, the software moves from continuous analysis to executing energy-saving recommendations across vendor-agnostic equipment and multi-site fleets.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

Data Centers

• [Link] Google announced plans to invest €13 billion ($15.1 billion) into AI infrastructure in Finland across 2027 and 2028, marking the company’s largest single investment in Europe to date. The funds will expand data centers and supporting facilities across four Finnish cities to boost computing capacity for services like Gemini, Search, and Maps. As part of the expansion, Google also secured clean energy supply through a 22-year nuclear power purchase agreement with Finnish operator Fortum along with new wind and battery storage projects.

• [Link] TikTok plans to secure 150 megawatts of capacity at a new data center facility in Kouvola, Finland, as part of its €12 billion “Project Clover” initiative aimed at strengthening data protection for its European users.

• [Link] Former Google executive Benjamin Treynor Sloss has launched Crux AI, a U.S.-based infrastructure company aimed at delivering dedicated, industrial-scale AI compute to enterprises and governments. Supported by a $5 billion equity commitment from Blackstone and technology from Google, the company plans to bring its first 500 MW of TPU capacity online in 2027 as part of a multi-gigawatt scaling strategy.

• [Link] Blue Owl Capital is reportedly evaluating a publicly traded data-center REIT, anchored by approximately $6.5 billion of its existing data-center assets.

• [Link] Palantir and Nebius have formed a strategic partnership designating Nebius as Palantir’s preferred sovereign AI infrastructure partner, bringing Nebius’s AI-native compute and inference capabilities directly into Palantir’s perimeter so enterprise customers retain full control over their data, models, and compute.

• [Link] The U.S. Department of Defense’s Office of Strategic Capital is in talks to lend roughly $5 billion to AI cloud computing startup Fluidstack to bolster domestic manufacturing and supply chain capacity for critical data center infrastructure, such as power delivery and cooling equipment.

• [Link] Singapore-based DayOne Data Centers launched Singapore’s first Biological Data Center Prototype in collaboration with Cortical Labs and the Yong Loo Lin School of Medicine, National University of Singapore (“NUS Medicine”). The deployment of a 20-unit CL1 biological computing system – the first independently operated biologically integrated server rack in the world – brings Cortical Labs’ technology into a live research environment at NUS Medicine, hosted within infrastructure designed and supported by DayOne.

• [Link] SemiAnalyis: What is So Hard About Behind-The-Meter Power For Datacenters? Part 1. Microsoft signs over 5GW of behind-the-meter nameplate capacity, Google deploys 930MW of off-grid aeroderivative turbines, and OpenAI will imminently start operations in its flagship off-grid 1.4GW (IT capacity) campus in Shackelford County, TX, using over five hundred 4.25MW Jenbacher J624 engines. The grid simply cannot keep up with demand.

• [Link] YouTube - SemiAnalyis: Modular Data Centers Cut Build Time to 12 Months

• [Link] Firmus has entered into a multi-year strategic partnership with OpenAI, which will serve as an anchor customer by contracting dedicated AI compute capacity from two new data center sites in Malaysia. This expansion into its fourth Asia-Pacific market brings Firmus’ total contracted capacity across all customers to more than 900 MW.

Space

• [Link] Despite US pressure encouraging major American space firms to skip the event, the International Space Summit in Paris highlighted the growing strength and independence of Europe’s space ecosystem. The summit yielded billions of Euros in major deals—including satellite build contracts for Eutelsat’s IRIS² constellation, mega-fundraises like The Exploration Company’s $450M Series C, and a flood of Ariane 6 launch orders from global partners. Overall, the gathering underscored Europe’s accelerating push for sovereign launch capabilities, mega-constellations, and global competitiveness in commercial space.

• [Link] French President Emmanuel Macron announced a $1 billion program led by Loft Orbital and Marlan Space to build a 50-satellite constellation, establishing an unprecedented artificial intelligence infrastructure in space. Partnering with companies like Mistral AI and BlackSky, the initiative enables satellites to process imagery and run agentic AI models directly in orbit for real-time applications like disaster response and critical infrastructure monitoring

• [Link] Lonestar takes its first step toward the New York Stock Exchange.

• [Link] Antares announced it has been selected by the Office of the Assistant Secretary of the Air Force for Space Acquisition and Integration for a Strategic Breakthrough award to advance nuclear power for space.

• [Link] The Quantum Threat to GPS, Defense & Satellites. Space Capital Founder and CEO Chad Anderson sits down with Eddie Zervigon, CEO of Quantum Exchange, to discuss why satellites are uniquely exposed to the emerging quantum threat.

[Link] Arm has launched the Arm Total Design for Physical AI ecosystem alongside a new Robotics Capability Framework to reduce engineering fragmentation and accelerate the development of autonomous systems. The framework establishes a shared six-level industry standard to classify, evaluate, and compare intelligent robotics across varying degrees of operational complexity. Meanwhile, the expanded partner ecosystem brings together hardware, software, and AI leaders to streamline integration across sensors, compute silicon, digital twins, and AI models.

[Link] GlobalFoundries has outlined parallel technology roadmaps for optical connectivity and power delivery to address scaling constraints in AI data centers. The company is advancing high-density silicon photonics for faster optical interconnects alongside GaN-based power architectures and integrated voltage regulation to deliver power more efficiently to processors.

[Link] Positron AI Raises $875 Million at a $5 Billion Valuation to Bring Its Next-Generation Inference Silicon to Market.

Energy

• [Link] NextEra Energy has finalized a $1.9 billion federal loan from the U.S. Department of Energy to fund the restart of the 615-megawatt Duane Arnold nuclear power plant in Linn County, Iowa (YIMBY), which shut down in 2020. Subject to regulatory approvals, the utility aims to bring the facility back online by early 2029 under a long-term power purchase agreement primarily aimed at supplying clean energy to Google’s artificial intelligence data centers.

• [Link] Google, Xcel Energy, utility companies, and clean energy trade groups are expressing support at FERC for the Midcontinent Independent System Operator’s (MISO) proposed “Zero-Injection” framework. The proposal creates a fast-tracked 90-day review process to interconnect power generation dedicated to serving large, co-located loads (such as data centers and crypto mines) that do not export power onto the broader grid.

• [Link] Toshiba and DIMAAG have partnered to integrate Toshiba’s high-power SCiB lithium titanium oxide (LTO) battery cells into DIMAAG’s immersion-cooled ZettaWatt Power Platform. This collaboration aims to deliver scalable, high-duty-cycle energy storage designed to stabilize power grids, smooth rapid load fluctuations, and provide reliable backup power for AI data centers.

• [Link] South Korea is finalizing an energy investment package worth over $100 billion with the United States to fulfill commitments made in a trade deal to reduce tariffs. The plan allocates funding toward constructing up to eight nuclear power reactors and developing natural gas infrastructure, including a $22 billion gas project in Texas.

[Link] IonQ has introduced Superion 256, its sixth-generation quantum computing platform designed to scale to fault-tolerant quantum computing through manufacturable, semiconductor-based Electronic Qubit Control. Built jointly with SkyWater for standard data center integration, the platform is available to order now, with initial customer deliveries planned for 2027.

[Link] Aegiq has integrated NVIDIA’s open-source Ising AI calibration models into its Artemis photonic quantum computer to automate the complex, daily tuning of its quantum hardware. Quantum processors traditionally require manual adjustments by specialized engineers to maintain stability and performance, which creates a major bottleneck as systems scale. By using this AI-driven process to automatically balance critical hardware parameters—such as photon brightness, purity, and indistinguishability—Aegiq achieved a 3x reduction in weekly engineering overhead while allowing non-specialists to manage routine maintenance.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

What I watched….

• [Link] SemiAnalysis and High Yield: Why AI is running out of Power. Deep dive into data center power demand, behind-the-meter demand, and how AI affects industries.

• [Link] Moonshots podcast. Hands down my favorite podcast

• [Link] Leo Cui: The Entire Space Data Center Explained

• [Link] Tesla: Deploying Megapack at scale. Learn how Megapack goes from the factory through installation and commissioning to get connected to the grid.

• [Link] Yet another awesome video from Anastasi in Tech: The new $119B Chip Factory Threatens TSMC. Terafab

Data Centers

• [Link] Ground Level AI. John Deere Moving Dirt for Data Centers. Although I found this post through Rich Miller, I’m going to claim it, since I live in John Deere country. I also worked directly with John Deere for ~5 years (a long time ago). John Deere is experiencing an unexpected growth boom driven by the surge in artificial intelligence data center construction. Strong demand for earth-moving machinery to clear land for massive data centers, infrastructure, and energy projects has boosted the company’s construction equipment sales by 5% to 10%.

• [Link] AMD, Cisco, and Saudi AI company HUMAIN have launched a production deployment of AMD- and Cisco-powered AI infrastructure in Saudi Arabia to deliver GPU-as-a-service for advanced AI workloads. Building on this live deployment, the joint venture plans to expand the Kingdom’s sovereign AI platform to 250 megawatts starting in 2027, staying on track toward a goal of up to 1 gigawatt by 2030.

• [Link] Nscale has announced a multi-year strategic partnership with humanoid robotics company Figure, becoming its preferred compute provider to power Figure’s Helix AI models and robots. The deal includes a multi-billion dollar compute commitment starting in Texas in 2027 and a strategic investment from Nscale into Figure. Nscale also closed on approximately $3 billion delayed draw term loan facilities.

• [Link] Sta’O’Nuk: How a new transpacific subsea cable will reshape cloud connectivity between North America and Japan. AWS is developing “Sta’O’Nuk,” a new 420 Tbps transpacific subsea fiber optic cable connecting Washington state and Japan, set to become operational in 2029 to boost cloud network capacity, speed, and resilience.

• [Link] Anthropic has signed a cloud capacity agreement with neocloud Lambda to the tune of $35 billion. As reported by the Wall Street Journal, and citing sources familiar with the matter, the agreement pertains to 350MW of capacity at a data center under development in Nueces County, Texas.

• [Link] Elon Musk is bringing turbine blade manufacturing in-house at SpaceX to eliminate severe supply chain bottlenecks for natural gas power generation. By casting these complex, high-temperature components in Texas, he aims to shorten multi-year lead times by up to 18 months, securing the massive electricity needed to rapidly power xAI’s data centers.

Space

• [Link] September 8th: Star Trek turns 60!

• [Link] KSAT and Unibap Space Solutions are collaborating to integrate Unibap’s iX20 edge computing platform into KSAT’s upcoming Hyperion demonstration mission, slated for launch in late 2027. This partnership combines low-latency satellite connectivity with onboard AI and high-performance processing to convert raw sensor data into actionable insights directly in orbit.

• [Link] The NASA Deep Space Network (DSN), best known for communicating with distant spacecraft such as Voyager 1, Voyager 2 and New Horizons, added a 114-foot (34-meter) radio-frequency antenna in California, boosting capacity on the increasingly crowded network. The DSN operates three complexes - at Goldstone in California, near Madrid in Spain, and Canberra in Australia - each of which includes a large 230-ft (70m) antenna alongside several smaller dishes.

• [Link] Blue Origin has won a NASA contract worth up to $700 million to build and operate a Mars Telecommunications Orbiter using its Blue Ring space platform. Scheduled to be operational by 2030, the mission aims to upgrade NASA’s aging Martian relay infrastructure to support upcoming robotic and human exploration.

• [Link] Cowboy Space has leased a 291,000-square-foot facility in Kent, Washington, to manufacture heavy-lift rockets and hardware for its planned constellation of solar-powered orbital AI data centers. The massive new production hub is expected to create 300 jobs as the startup expands its space manufacturing operations in the Seattle area.

• [Link] Sparkle and Hellas Sat successfully demonstrated the first quantum-safe satellite connection between Greece and Cyprus, using Sparkle’s software-based Quantum Safe Interconnect platform over a 72,000-kilometer geostationary link to protect data against quantum computing threats.

• [Link] Kongsberg Satellite Services (KSAT) executive Edvard Foss details the company’s plans to expand its global ground network into space using HYPER, a next-generation hybrid radio frequency and optical in-orbit relay constellation. Operating as “orbiting ground stations” in Low Earth Orbit, the network uses both Ka-band/S-band RF and optical laser links - tested via its Hyperion pathfinder mission launching in late 2027 - to deliver secure, low-latency data relay and real-time satellite tasking for commercial and institutional customers.

Chips

• [Link] Fabrication platform could enable flexible, transparent next-generation photonic chips. MIT researchers, collaborating with NY Creates, have developed a scalable wafer-scale manufacturing process to produce silicon-photonics chips that are both mechanically flexible and optically transparent. By utilizing standard 300-millimeter semiconductor tools, the platform preserves optical performance even after thousands of bends down to millimeter radii. This breakthrough paves the way for commercial-scale production of advanced photonics applications, including curved augmented-reality visors and discreet, wearable health monitors.

• [Link] Silicon photonics company iPronics raised $125 million in Series B funding, with participation from NVIDIA, Maverick Silicon, and Light Street Capital, to scale commercial deployment of its programmable optical networking platform for AI data centers.

• [Link] Architect Labs introduced Redwood, a low-power frontier AI accelerator whose hardware design, verification code, and software stack were generated almost entirely by an AI system in under two weeks to run edge inference workloads.

• [Link] One to watch: Source Foundry is a Stanford-founded semiconductor equipment startup building next-generation lithography and chip-fabrication tools to break the ASML bottleneck, with the potential to unlock cheaper, faster AI chip production and reshape the global advanced-node supply chain.

Energy

• [Link] Google has partnered with Fervo Energy on a historic power purchase agreement for 396 megawatts of enhanced geothermal power from Utah’s Cape Station, with an option to scale up to nearly 1 gigawatt by 2030. This clean, around-the-clock electricity will serve as a foundational energy source for Google’s potential future data center presence in the state.

• [Link] SMA Solar Technology has expanded into the data center market by launching a portfolio of power infrastructure solutions designed to accelerate time-to-power for AI and hyperscale applications. Built on its utility-scale energy technology, the new solution architectures - SMA GridAssist, SMA GridLink AC, and SMA GridLink DC - aim to solve grid interconnection bottlenecks and provide reliable, scalable power.

• [Link] NeoVolta Power has entered a five-year agreement with South Korean battery manufacturer SK On to secure 9 gigawatt-hours of U.S.-manufactured LFP battery cells from 2027 through 2031. The collaboration also establishes a framework for SK On to supply an additional 9 GWh of cells while purchasing energy storage packs manufactured by NeoVolta Power, expanding the potential activity between the companies up to 18 GWh.

• [Link] Emerald AI, Alderbuck Energy, and UC San Diego have partnered to pilot a grid-responsive power architecture for AI data centers at the San Diego Supercomputer Center. The project integrates Alderbuck’s solid-state transformer hardware with Emerald AI’s workload orchestration software to deliver 800V DC power directly while dynamically adjusting AI compute tasks to support grid reliability and reduce energy waste.

• [Link] Google has selected 28 startups across North America and Europe to participate in its Google for Startups Accelerator program focused on AI-driven energy solutions. Over the coming months, these companies will receive technical mentorship and access to Google’s digital tools to help develop technologies that modernize power grids, lower costs, and enhance energy efficiency. Gridstatus.io stands out as a top resource (to me), featuring data-dense dashboards and powerful apps designed for real-time energy tracking.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

[Link] Nordic AI infrastructure company Fossefall has selected Armada as one of its technology partners for future AI factories in Norway, Sweden, and Finland. The first order of five Leviathan modular data centers has been placed as part of Fossefall’s plan to build more than 1 gigawatt of Nordic-owned AI capacity by 2030.

[Link] SK Telecom is spinning off the data center and submarine cable operations of SK Broadband into a new AI data center infrastructure company called SK Horizon. To fund its expansion, global investment firm KKR and the IMM Investment-Stonebridge consortium are together investing KRW 3.08 trillion (roughly $2.241 billion) for a combined 49% stake in the new entity.

[Link] As a follow-up to the Hot Interconnects stories last week, Alex Woodie from HPC Wire has a nice piece on the next generation of optical interconnects. You know, the acronym soup of OCI, Open CPX, SDM4 MCF, and XPO.

[Link] Photonic Inc.’s paper on SHYPS quantum error correction codes, “Computing Efficiently in QLDPC Codes,” has been published in the journal Nature Communications. Quantum Low Density Parity Check (QLDPC) codes allow larger programs to be run on smaller systems, accelerating the timeline to commercially relevant quantum computing.

[Link] Renesas Electronics has launched a new Physical AI & Robotics Lab in Beijing to serve as a hub for validating and deploying next-generation robotic systems. The dedicated engineering facility combines hardware, software, sensing, and AI modeling to help Renesas expand its component coverage in humanoid robotics.

[Link] The Machine Age Fund. a16z launched the Machine Age Fund, a $1.1 billion venture fund dedicated to accelerating the physical buildout of artificial intelligence. The fund focuses on investing in the hardware and critical infrastructure necessary to scale AI, including semiconductors, data centers, networking, storage, robotics, and consumer hardware.

[Link] Anthropic has introduced a research preview of the Model Hardware Standard (MHS), an open-spec framework developed with HHMI Janelia that allows AI agents to safely discover, control, and orchestrate physical equipment like lab instruments and robotic arms. Using standardized drivers and commands, MHS reduces device integration times from months to minutes and enables real-time, autonomous experimental workflows across disparate hardware.

[Link] SK hynix and a team of global academic researchers published a comprehensive roadmap for Co-Packaged Optics (CPO) in Nature Electronics to address the “bandwidth wall” in hyperscale AI systems. The proposed optics-centric architecture replaces traditional copper interconnects with integrated optical links, allowing memory and processors to communicate using light. This transition drastically improves data-transfer bandwidth, energy efficiency, and scalability across large-scale AI data centers.

[Link] Skyloom Global, an IonQ company, announced that dozens of additional optical communication terminals (OCTs) have been successfully deployed on orbit. The OCTs were aboard York Space Systems satellites supporting the Space Development Agency’s (SDA) Proliferated Warfighter Space Architecture (PWSA). IonQ now has 84 Skyloom optical communication terminals installed in orbit.

[Link] AWS and NVIDIA to Deliver 2 Million Additional GPUs and Next-Generation Infrastructure for Agentic and Physical AI. Building on already rapid customer adoption of NVIDIA-accelerated compute on AWS, the companies plan to deploy 2 million additional NVIDIA GPUs across AWS’s global infrastructure and deepen their work together across AI factories, CPUs, networking, open models, data processing and robotics, delivering co-engineered AI solutions.

[Link] DE-CIX Chicago has reached a major connectivity milestone by exceeding 1 Terabit per second (Tbps) of peak network traffic for the first time

[Link] In The Interconnection Fast Pass: Can a $17M Battery Really Replace 5 Years in the Queue?, Christopher Johnson explores whether data center developers can bypass multi-year grid queue delays by investing heavily in battery storage infrastructure. Using Aligned Data Centers as a primary case study, the article analyzes how a upfront $17 million battery investment can serve as a work-around for grid constraints to accelerate project timelines.

[Link] Episode 280 of the Moonshots podcast: 200GW Hiding in Grid, Sodium Batteries 10x Cheaper, Wave-Powered Datacenters w/ Ramez Naam. My favorite podcast hosted Ramez and discussed the state of energy, the grid’s struggle to keep pace with AI, breakthroughs in sodium batteries and fusion, and whether wave-powered data centers could unlock a new era of energy abundance.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

Data Centers

• [Link] NVIDIA is strategically securing land, power, and shell (LPS) capacity - a critical constraint for AI factories - to directly host its full-stack compute infrastructure for key partners like OpenAI. Through a long-term partnership with SB Energy at the PORTS-Pike Technology Campus in Ohio, NVIDIA will support 4.25 gigawatts of capacity over 20-year upgrade cycles to serve high-demand frontier AI labs. This initiative allows NVIDIA to lock in long-term, multi-generation compute upgrades, representing an estimated $600 billion opportunity with OpenAI through 2030.

• [Link] Cloverleaf Infrastructure announced it has entered into a strategic partnership with NVIDIA to advance digital infrastructure development across the United States. As part of the partnership, NVIDIA made a minority investment in Cloverleaf to further support the build out of AI factories.

• [Link] Google and Beale Infrastructure have announced a new 506-acre data center campus under construction in Tulsa County, Oklahoma. The project includes major investments in local utility systems, water line installations, road upgrades, and electrical workforce training to strengthen regional infrastructure.

• [Link] Firmus Technologies has agreed to acquire Benmax’s fabrication, design, and projects businesses for A$300 million to expand its vertically integrated operations and accelerate the deployment of its modular AI Factory platform.

• [Link] Backed by up to $1 billion from I Squared Capital, Saragon launches with ten data centers across nine major U.S. markets, 53 MW of installed power capacity and significant room for expansion. Saragon was established following I Squared’s acquisition of ten data center facilities from Cogent Fiber.

• [Link] Mousterian Corporation and Samsung Heavy Industries have signed an engineering contract to build moored, floating data centers designed for AI workloads in the United States, with initial deployments planned for Texas. Each facility will provide 50 megawatts of critical IT capacity, utilizing shipyard fabrication to significantly accelerate construction timelines compared to traditional land-based facilities. The floating units will feature zero-potable-water non-evaporative cooling and reduced noise profiles, aiming for commercial operation by 2028.

• [Link] Data Center Richness: Ron Vokoun: Navigating Change at the Speed of AI. Rich interviews industry veteran Ron Vokoun, who shares his insights on modular prefabrication, the rapid shift to liquid cooling, and how physical constraints are helping prevent an AI market crash.

Infrastructure-Scale Capital

• [Link] Starcloud has raised $250 million to allow the company to open a larger manufacturing facility and advance its largest orbital data center spacecraft, Starcloud-3, which is intended to fly on SpaceX’s forthcoming Starship rocket.

• [Link] Muon Space has closed an oversubscribed $250 million Series C funding round - bringing its total equity funding to over $386 million - to expand its Silicon Valley satellite manufacturing capacity and accelerate the deployment of turnkey space infrastructure for commercial and government clients.

• [Link] Broadcom is reportedly seeking to borrow up to $100 billion as part of a new artificial intelligence chip financing deal.

• [Link] Vantage Data Centers is exploring strategic options including a potential sale or an IPO as soon as next year that could raise around $10 billion and value the company at approximately $100 billion.

• [Link] Nebius announces pricing of upsized private offering of $5.0 billion.

• [Link] Groq announced a $350 million Series A fundraise and the round was led by Disruptive, with planned participation from NVIDIA.

• [Link] ST Telemedia Global Data Centres announced that it has secured a green financing facility of up to $1.37 billion (MYR 5.6 billion) to support the development of STT Johor, its flagship Malaysia data centre campus at Nusa Cemerlang Industrial Park in Johor.

• [Link] Frontier inference hardware company Etched announced it has shipped its first rack to Jane Street, and raised $700M at a $21B valuation.

• [Link] NextEra Energy has signed agreements with the US Department of Commerce and the Government of Japan to fund up to 10GW of new natural gas generation across Texas and Pennsylvania. The deal will release $3.3 billion in funding for two projects, including one in Texas and the $17bn South Mon project in Pennsylvania.

• [Link] AI infrastructure company Velaura announced that it has raised $110 million in Series A financing, bringing its total valuation to more than $1 billion.

Hot Interconnects 2026: Scale-Up, Scale-Out, Scale-Across: Do they really differ?

• [Link] In his IEEE Hot Interconnects keynote highlighted by HPCwire, Meta Senior Director of Networking Omar Baldonado emphasized that scaling gigawatt-class AI fleets requires managing three distinct interconnect domains: scale-up within racks, scale-out across a data center, and scale-across between multi-building campuses. He noted that the primary challenge of scale-across networking is handling hardware and workload heterogeneity across vast distances rather than relying solely on raw bandwidth.

• [Link] Lots of news on X for the upcoming Hot Chips 2026 conference. #HotChips2026

• [Link] Very nice infographic from John Furrier on the AI chip war.

Quantum

• [Link] Researchers from the State University of New York at Stony Brook University and the U.S. DOE Brookhaven National Laboratory successfully transmitted quantum information - including entangled photons - through open air over a 13-mile free-space optical link between their facilities. This milestone adds a wireless capability to the nation’s longest quantum network, laying the foundation to connect quantum devices beyond the limitations of fiber-optic cables.

• [Link] IBM has successfully connected and cooled down two modular cryogenic systems into a single ultra-cold environment. This achievement marks a key engineering milestone toward linking hundreds of quantum processors, supporting IBM’s goal to launch its first fault-tolerant quantum computer by 2029.

• [Link] NTT DOCOMO has deployed its second D-Wave-powered quantum application in live network operations to optimize mobile Tracking Area Lists across multiple tracking areas. The deployment reduced peak location registration signals by 65.3% and paging signals by 7.0%, significantly easing network signaling loads and boosting operational efficiency.

Energy

• [Link] SemiAnalysis: PJM’s $12B Modeling Mistake Is Hitting Ratepayers Again (Datacenter, Energy). Robert Boswall (@RobertBoswall) and Jordan Nanos (@JordanNanos) break down how the largest grid in America, 13 states and 66 million people, inflates demand while constraining supply.

• [Link] Bloom Energy has introduced Power Connect, a new standardized deployment system designed to reduce on-site power installation time by over 40%. By shifting electrical integration from construction sites to factory assembly, the pre-wired equipment helps power-intensive operations like data centers bring energy capacity online faster and more reliably.

[Link] Zayo Group has expanded its strategic supply agreement with Corning Incorporated to secure a major portion of the fiber optic cable needed for its long-term network growth. This deal ensures reliable material supply as Zayo executes a massive expansion to add 15,000 route miles of fiber by 2030 to support surging AI infrastructure demand.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

Data Centers

• [Link] In this entry of a16z’s Charts of the Week series, author Moses Sternstein highlights the rapid rise and financial exposure of “Neocloud” AI infrastructure providers such as CoreWeave, Nebius, and Iren. While these specialized cloud platforms are seeing rapid sales growth, their total liabilities and financial obligations (driven by debt and hardware leases) are outpacing revenue, with capex-to-sales ratios reaching unsustainable levels. The edition also touches on broader software and market trends, including the resurgence of horizontal SaaS, efficient token-optimization strategies, and intensified competition for AI lab talent.

• [Link] Tech Titans Outline Their Vision for 800 VDC. Major tech leaders - including NVIDIA, Google, Microsoft, and the Open Compute Project (OCP), have collaborated on a new specification for solid-state transformers to advance high-voltage 800 VDC power distribution in next-generation AI data centers.

• [Link] Fermi has signed its first binding customer lease with AI cloud provider TensorWave for a 222 MW data center facility at its Project Matador campus in Texas, expected to generate approximately $6.5 billion in contracted revenue over a 15-year initial term starting in late 2027.

• [Link] According to updates shared during a recent Alliant Energy earnings call, several marquee data center projects involving Google, QTS, and Meta are actively progressing under full construction. Google has already energized transmission service in Cedar Rapids, QTS’s seven-building campus there is slated for initial 300 MW energization later this year, and Meta’s Beaver Dam site is undergoing vertical construction. Beyond these current builds, Alliant has signed a new 370 MW agreement and intends to file a 900 MW interconnection request for a second Iowa project, deepening its focus on hyperscale growth and expanding its long-term load trajectory.

• [Link] Infrastructure Masons: The Next Data Centers: Ocean Floor, Job Site, Space

• [Link] InfraPartners and Phaidra have partnered to integrate Phaidra’s autonomous AI agents - including its liquid-cooling and operational intelligence software - into InfraPartners’ prefabricated AI data centers. This collaboration aims to optimize real-time cooling controls, streamline facility management, and safely maximize the power available for high-density compute workloads.

Space

• [Link] The U.S. Space Force has awarded $12 million contract and agreement sets to five companies (Amazon LEO for Government, Lockheed Martin, Northrop Grumman, Rocket Lab and York Space Systems) to help build a resilient, open-architecture Space Data Network. By establishing standardized physical, electrical, and data interfaces, the program enables diverse commercial satellites to seamlessly connect and stream data without locking the military into single-source vendors.

• [Link] Intel Patent Describes Orbital Data Centers Designed to Manage Satellite Networks. Intel has filed a patent (US 2026/0230175 A1) proposing data centers built into satellites in medium, geosynchronous, or high elliptical orbit - designed to manage and process data for entire low-Earth orbit constellations. The core idea: instead of LEO satellites bouncing data down to ground stations for processing, higher-orbit satellites would act as a “control tower,” handling compute, network routing, and inter-satellite coordination for the LEO layer below - cutting out the round trip to Earth.

• [Link] Everyday Astronaut: Inside Starlab, a MASSIVE commercial Space Station! Tim goes inside Starlab by Voyager Technologies with their CEO Dylan Taylor.

• [Link] Voyager Technologies was awarded a contract by the U.S. Space Force Space Systems Command to develop space-to-space communication capabilities for a resilient Department of Defense data transport architecture. As part of the project, Voyager will design a flight-ready communications waveform for multiple orbits and execute an on-orbit demonstration.

• [Link] Houston-based Intuitive Machines, The Space Infrastructure company, ended the second quarter with a record backlog of $1.8 billion, an increase of $1.5 billion from year-end 2025 as the company booked record levels of diverse awards across commercial, civil, and national security space customers.

• [Link] I liked this opinion piece by Mathieu Luinaud that argues that satellite constellations have shifted from basic communications infrastructure into critical geopolitical and economic assets, driven by the integration of AI.

[Link] Lightmatter and a 19-company industry coalition have formally launched the “Open Silicon Photonics for AI Systems“ workstream within the Open Compute Project to create standard co-packaged optics (CPO) rack architectures. The group released a foundational 300-page white paper establishing a technology-agnostic framework to ensure multi-vendor interoperability and scale optical interconnects across massive AI clusters.

[Link] Atomica Launches AI Photonic Integration Platform. Atomica’s AI Photonic Integration Platform addresses a critical bottleneck in AI hardware: integrating optics, electronics, memory, thermal management, and packaging into one manufacturable system. The platform gives customers a configurable microfabrication pathway spanning interposers and vias, bonding, optical alignment, and wafer-level packaging.

[Link] Here’s another one I’ll pretend to fully grasp, but honestly I’m mostly just sharing the link because I found it fascinating: Free Electron Lasers and the Birth of the Beamline Tycoon. Elon also dropped a six-letter post recently, FELFTW (Free Electron Lasers For The Win)... right around when the Terafab announcement dropped.

[Link] Quantinuum and Quanta Computer have entered into a strategic partnership to co-develop hardware infrastructure, systems engineering, and manufacturing capabilities for future generations of quantum hardware. The collaboration aims to transition Quantinuum’s trapped-ion quantum technology from lab-scale prototypes into scalable, modular, and commercially deployable fault-tolerant systems.

[Link] Qunnect announced a contract from DARPA to advance the next generation of its Carina quantum entanglement distribution platform, focusing on enhancing its polarization compensation module. The upgraded technology aims to continuously correct for environmental interference and improve the reliability of quantum networks operating over existing telecommunications fiber.

[Link] Infleqtion has been selected by Eaton for a subcontract under an Air Force Research Laboratory award to apply quantum computing to U.S. electrical grid contingency analysis. The multi-year research program aims to leverage quantum hardware and algorithms to predict and prevent power outages, evaluating failure scenarios faster and more accurately than classical systems.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

[Responsible] Data Centers. Less noise, more signal. The data center industry is packed with incredible innovation and untapped potential, but it’s easy to get lost in the recent news cycle. I try to cut through the static to deliver constructive, forward-focused coverage on where the industry is actually heading. …and yes, I had AI help me write that.

• [Link] University of Iowa data center decision tool. The newly launched Iowa Data Center Decision Tool for Counties, Cities, Communities provides local leaders with an objective framework designed to help local governments weigh big-tech development against community priorities. Built by and for Iowans, the platform equips city and county officials with key questions to tackle vital topics - like water usage, energy demand, and zoning - long before signing contracts or approving major proposals.

• [Link] A bipartisan coalition of Iowa labor, business, and economic development leaders has launched Growing Iowa’s Economy, an initiative advocating for the responsible development of data centers and digital infrastructure in the state. The group aims to attract billions of dollars in investments, create thousands of jobs, and boost local property tax revenues, all while ensuring data center operators pay for their own energy consumption and grid upgrades.

• [Link] Chris Crosby: Communities deserve better partners, not blanket bans. Chris outlines the principles he believes communities should expect from every data center developer, including responsible infrastructure investment, transparent communication, thoughtful environmental stewardship and long-term commitment to the communities where they operate.

• [Link] Texas Governor Greg Abbott directed the Public Utility Commission of Texas and ERCOT to perform a comprehensive audit and verification of all data center projects seeking grid interconnection, pausing further project approvals until compliance with power, water, financial, and community impact requirements is established. It was noted that the audit could delay 49.8 GW of new data center electricity demand from advancing on the ERCOT system - nearly 20% of the U.S. development pipeline - BloombergNEF (BNEF) said in an Aug. 5 report.

Data Centers

• [Link] Volta emerged from stealth last week with a mission to build The Utility of Compute. They hit the ground running, with a $10 billion deal with Anthropic. Volta has a partner in this deal, Bitdeer, a crypto-mining company that will help develop the data center to provide the compute capacity. That facility will be located in Norway and will deliver a 133 MW capacity. The 7-month-old company launched a $5 billion AI infrastructure program, and has a $2.4 billion valuation.

• [Link] Australian AI infrastructure provider Firmus has raised a fully subscribed $2 billion strategic equity round from investors including NVIDIA, Coatue, Blackstone, and Jane Street, bringing its valuation to over $10.5 billion.

• [Link] On August 3, KKR announced the final closing of its $19.2 billion flagship core-plus fund, KKR Global Infrastructure Investors V, aimed at investing in critical infrastructure assets across North America and Western Europe.

Energy

• [Link] Bloom Energy has expanded its partnership with MiTAC Computing Technology to deploy a fuel cell microgrid at MiTAC’s AI server manufacturing campus in Fremont, California. This agreement increases Bloom’s contracted on-site power capacity for MiTAC and reinforces its growing footprint in powering AI infrastructure.

• [Link] AVK SEG has secured its first-ever external investment, with private equity firm Partners Group acquiring a majority stake in the company with an initial commitment of over $1 billion. This funding will support AVK’s growth as a European leader in power solutions for data centers and AI infrastructure, specifically driving the expansion of its Energy-as-a-Service (EaaS) model to build, fund, and operate on-site microgrids. The company recently launched modular PowerPods for power delivery built around infrastructure repeatability.

• [Link] Energy Vault has executed a strategic commercial agreement to supply battery energy storage systems (BESS), power conversion systems, and AI control software for a 1.25 GW off-grid power infrastructure project serving a hyperscaler AI data center in Texas. The partnership incorporates Caterpillar gensets and turnkey power generation to bypass traditional grid connection delays, generating an estimated $500–$600 million in revenue for Energy Vault across late 2026 and 2027.

Physical AI

• [Link] Making a Billion Intelligent Machines. Marc Andreessen describes Applied Intuition’s launch of Dana, an agentic platform that layers AI agents on top of the company’s long-built stack for developing intelligent physical machines.

• [Link] NVIDIA has announced the availability of Alpamayo 2 Super, a 34-billion-parameter open vision-language-action (VLA) model designed for autonomous vehicle and robotaxi development. Built on NVIDIA’s Cosmos 3 backbone, the model processes 360-degree multi-camera video to generate trajectory predictions, causal reasoning explanations, and structured auto-labels, and is now released under an open commercial license.

• [Link] TDK and LG Innotek have entered into a strategic partnership to develop sensing technologies for physical AI applications, with an initial focus on next-generation vision and tactile sensing modules for humanoid robots. The collaboration combines LG Innotek’s expertise in optical systems and ultra-precision module design with TDK’s sensor technology portfolio to create integrated solutions that enhance environmental perception and interaction.

• [Link] AMD announced the AMD Robotics Partner Network (RPN), an open and structured partner ecosystem designed to accelerate the development, validation, and deployment of robotics and physical AI systems across embedded, edge, and data center environments.

• [Link] SK Telecom announced that the International Telecommunication Union (ITU-T) adopted its proposed standardization work item for a “Robot Data Factory” architecture. This initiative aims to establish global standards for systematically producing, exchanging, and utilizing data required to train physical AI and intelligent robots across diverse platforms.

Quantum Computing

• [Link] IonQ CEO Niccolo de Masi discusses Quantum Computing, Q-day and National Security. Really cool video. Disclosure: I own an insanely small amount of IonQ stock, and really like what the company is doing.

• [Link] IonQ has been awarded a $28 million DARPA contract extension—with an option for an additional $30 million—under the “It’s About Time” program to produce and deliver 125 next-generation Evergreen-05 optical atomic clocks to U.S. government customers for mission-critical defense applications.

• [Link] QuiX Quantum announced the commercial availability of Alquor 2.0, the next generation of its rack-mountable quantum photonic processor platform, available in 8-mode, 20-mode, and 32-mode configurations, for advanced quantum optics, quantum communication, boson sampling, quantum information processing, and photonic quantum computing research.

• [Link] D-Wave has announced a research breakthrough published in Nature, demonstrating a fast, high-fidelity two-qubit entangling gate using its superconducting dual-rail architecture. Built with native hardware-level error detection, the technology significantly reduces the physical qubit overhead needed for quantum error correction, advancing the path toward practical, fault-tolerant gate-model quantum computing.

Space

• [Link] Rocket Lab has been awarded a $397 million contract by the U.S. Space Force under the Space-Based Airborne Moving Target Indicator (SB-AMTI) program to develop, launch, and operate a constellation of next-generation satellites designed to track airborne threats from orbit. Under the award, Rocket Lab will build multiple Flatellites, its new flat satellite platform optimized for large constellations, and launch them aboard its upcoming Neutron rocket.

• [Link] Neuraspace has secured €15.6 million in combined private and public funding from investors and Portugal’s Recovery and Resilience Plan to accelerate its AI-driven space traffic management and defense-grade space domain awareness capabilities. The investment will expand the company’s proprietary optical sensing infrastructure, enhance autonomous satellite mission operations, and scale its dual-use platform for commercial and defense users.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

Data Centers

• [Link] SemiAnalysis: The Wild Wild West Of LEGO Datacenters. The SemiAnalysis Industrials model is a complete data center supply chain map, and their Modular Tracker, included in the model, tracks over 61GW of modular capacity and 1,000+ sites using some form of modularization or prefabrication strategy. Grab some popcorn and dig in to this 10,000+ word deep dive.

• [Link] The U.S. Department of Energy has partnered with a consortium led by Brookfield and NextEra Energy to develop a $100 billion, privately funded artificial intelligence data center campus at the former Paducah uranium enrichment site in Western Kentucky. Scheduled for full completion by 2032, the project will feature a 1.8-GW data center campus operated by Brookfield. To avoid passing infrastructure costs on to local ratepayers and to shield the existing grid, NextEra will build and own up to 4.6 GW of dedicated private power support.

  • [Link] On a recent post-earnings call, CEO John Ketchum announced that NextEra had acquired the remaining 30% stake in Duane Arnold - a nuclear power plant located in Palo, Iowa - from its cooperative partners, giving the company full ownership.

• [Link] Core Scientific and AMD announced a strategic partnership where AMD will secure more than 500 megawatts of U.S. data center capacity from Core Scientific beginning in 2027, with the option to expand up to 2.5 gigawatts.

• [Link] Meta and BlackRock announced a venture to develop and own a data center campus in El Paso, Texas. Funds managed by BlackRock will own an 80% interest in the venture, while Meta will retain the remaining 20% ownership. The parties have committed to fund their respective pro rata share of the approximately $14 billion in total development costs for the buildings and long-lived power, cooling, and connectivity infrastructure at the campus.

• [Link] Private equity firm TPG Inc. is in talks to acquire Netrality Data Centers from majority owner Macquarie Asset Management in a deal that could value the operator at up to $3 billion. The potential acquisition would give TPG control over Netrality’s portfolio of 18 US-based interconnection data centers amid surging demand for digital and AI infrastructure.

• [Link] Galaxy has purchased roughly 500 acres in McGregor, Texas, to build a new data center campus dedicated to AI and high-performance computing. Located inside the McGregor Industrial Park, this marks the company’s second major data center investment in the state, adding to its massive 1.6-gigawatt Helios campus in West Texas.

• [Link] Crusoe and modular nuclear reactor company Aalo Atomics announced a strategic partnership focused on developing and deploying the first nuclear-powered AI Factory data center designed to validate nuclear power’s effectiveness with AI workloads. Aalo will initially power a Crusoe Spark modular data center, running Crusoe Cloud, in 2027 at Idaho National Laboratory as a proof of concept.

Space

• ❗[Link] A few weeks ago I wrote about 021T Capital investing in Lonestar to help build the Sovereign Dyson Swarm. In his Substack, The Innermost Loop (one of my top 5 favorite), Lonestar adviser Dr. Alex Wissner-Gross takes it a step further and discusses the first orbital librarian. “Hermes, the open-source agent framework from Nous Research, will fly on its StarVault launch planned for early next year, as the first orbital librarian, an autonomous agent whose standing job is curating a sovereign knowledge collection without end. Lonestar’s Data Embassy was storage. Its Sovereign Model was mind. This is the librarian.”

• [Link] Sophia Space and Caltech have been issued a joint U.S. patent (12,679,564) for a modular, solar-powered space-based data center architecture that utilizes passive cooling to process data directly in orbit. In tandem with the patent issuance, Sophia Space is expanding its research and development partnership with Caltech to further develop the structural and thermal technologies required to scale these on-orbit computing platforms.

• [Link] Intel has introduced “Starfire,” a space-grade system-on-chip internally funded to run advanced artificial intelligence and data-processing workloads directly onboard spacecraft. Manufactured using the company’s 18A process, the radiation-hardened processor is currently undergoing extreme-environment testing and is expected to ship to customers later this year.

• [Link] NASA is testing its next-generation High Performance Spaceflight Computing (HPSC) processor, a radiation-hardened system-on-a-chip developed with Microchip Technology that offers up to 500 times the processing power of current spacecraft chips to enable autonomous, AI-driven deep-space missions.

• [Link] Satellite startup K2 Space has raised a $500 million Series D funding round at a $6.8 billion post-money valuation to scale its production capacity and expand its high-power orbital satellite capabilities amid growing commercial demand.

  I’m fascinated by space communications and am actively teaching myself more about the field…

• [Link] The U.S. Space Force awarded K2 Space a $22.9 million contract to test laser terminals designed to transmit data directly between satellites in orbit. K2 Space will procure, integrate, launch, and operate two Enterprise Space Terminals, with each terminal carried as a hosted payload on a separate K2 satellite.

• [Link] Pulse Space announced a $40M contract from the U.S. Space Force to develop their laser-based remote power and situational awareness systems for space.

• [Link] Italian deep-tech startup ORiS has closed a €4.5 million pre-seed round to develop laser-based wireless power transmission technology for space and dual-use applications.

• [Link] ALL.SPACE launched Hydra MAX, a next-generation wideband SATCOM terminal platform that provides resilient, simultaneous multi-orbit and multi-network connectivity across commercial and government Ka-band networks. Following the company’s acquisition by York Space Systems and its relocation to the United States, this platform delivers 500 MHz of instantaneous bandwidth per beam to ensure secure communications in heavily contested environments.

• [Link] SES Space & Defense and Starlab have agreed to collaborate on communication architecture to deliver continuous, resilient, low-latency connectivity services supporting the full lifecycle of Starlab’s commercial space station via its LEO Relay Services (LRS). Starlab, which is developing the commercial space station under NASA’s Commercial Low Earth Orbit Development (CLD) Program, selected SES Space & Defense due to its integrated multi-orbit space and ground architecture.

• [Link] Space traffic intelligence as a new layer of critical infrastructure. Disclosure: I am a big fan of Privateer, but primarily because I am such a big fan of co-founder Steve Wozniak. This is a really good blog post about space traffic intelligence. Space situational awareness (SSA) is the ability to detect, track, and catalog objects in orbit. Space domain awareness (SDA) goes further: it adds the analytical layer that interprets what those objects are doing and why. Space traffic intelligence is the operational output of that work, the usable understanding that lets a government agency, satellite operator, or commercial team make a decision.

• [Link] Telespazio inaugurates a new S/X-band satellite antenna in the Azores, expanding its international network of ground stations. From my research: When a ground station or satellite antenna is described as S/X-band, it means the infrastructure is equipped with dual-capability hardware. It can use the highly reliable S-band to send vital commands and track a satellite’s position, while simultaneously using the high-bandwidth X-band to download large volumes of science data, imagery, or mission payloads.

[Link] Four leading quantum companies have selected New Mexico’s ABQ-Net, the only open-access entanglement-based quantum network in the United States, to develop and validate next-generation security technologies. The four companies are Infleqtion, Bandelier Technologies, Tensora, and Aliro Technologies.

[Link] Extropic has signed a non-binding $75 million letter of intent with the U.S. Department of Commerce under the CHIPS R&D program to scale and onshore the manufacturing of its Thermodynamic Sampling Units (TSUs). These specialized, energy-efficient chips leverage thermodynamic noise for probabilistic AI workloads and can be fabricated on mature nodes at domestic foundries.

[Link] The Open Programmable Infrastructure (OPI) Project has announced its first coordinated release, “OPI Abstraction v0.1.0,” which establishes a vendor-neutral, hardware-agnostic API layer across 26 repositories to standardize DPU (Data Processing Units)and IPU (Infrastructure Processing Unit) ecosystems. This release introduces “OPI Blueprints”—beginning with a Kubernetes Network Function Offload production pattern, to reduce operational complexity, eliminate vendor lock-in, and accelerate real-world deployments.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

Data Centers

• [Link] Nvidia in talks to provide $250B guarantee for 10GW OpenAI data center project in southern Ohio (WSJ).

• [Link] Project Camellia is a long-term datacenter project that OpenAI is designing and developing in Effingham County, Georgia. OpenAI will partner with Georgia Power Company for 3.2 gigawatts of power which will be delivered in phases between 2028 and 2032.

• [Link] Crusoe and ON.energy have partnered to deploy 5 gigawatts of ON.energy’s patented AI UPS technology across multiple hyperscale data center campuses, with commissioning beginning in 2026. This medium-voltage uninterruptible power system is designed to isolate local power grid infrastructure from rapid power demand fluctuations caused by synchronized GPU clusters during AI training.

• [Link] According to a report by The Information, SpaceX is planning to build or retrofit at least one new large-scale data center in Texas to significantly expand its AI computing capacity beyond its current Memphis-area facilities.

• [Link] Pure Data Centres Group (Pure DC) has secured €1.3 billion in senior project financing to fund the first phase of its fully leased, 110MW AI data center campus in Seinäjoki, Finland, bringing its total funding over the past year to more than $4.2 billion to accelerate hyperscale and AI infrastructure expansion across Europe and the Middle East.

• [Link] Rubix Data Centers, a subsidiary of Submer Group, has proposed building an $8-$12 billion, 2 GW hyperscale AI data center at a former AK Steel mill brownfield site in Russell, Eastern Kentucky.

Data Center Zeitgeist

• [Link] Google Trends data 2004-Present for “data center”. Getting harder to find the signal with all of this noise.

• [Link] Eric Schmidt says cash, not energy, is the real limit on AI growth

• [Link] New research from Synergy Research Group says AI demand pushes US data center capacity toward a doubling by 2029.

• [Link] Goldman Sachs raises data center forecast to 217 GW by 2030.

Quantum

• [Link] PsiQuantum Signs $125 Million Agreement with DARPA. PsiQuantum announced that the company has signed a new $125 million expanded agreement with the Defense Advanced Research Projects Agency (DARPA) under DARPA’s Quantum Benchmarking Initiative (QBI), the U.S. government’s most rigorous effort to evaluate commercial pathways to utility-scale quantum computing.

• [Link] Photonic Quantum Computing Companies. I’m going to bookmark this one. I am interested in the topic of photonic quantum computing. By that I mean, I would love to pretend to understand a fraction of what this technology is, and how it works. This Quantum Insider brief does a nice job of breaking down the topic, and then listing the companies that are in the space; like my favorite PsiQuantum.

[Link] Following a historic trade and investment agreement between the U.S. and Taiwan, Taiwan Semiconductor Manufacturing Company (TSMC) announced an additional $100 billion investment in American semiconductor facilities, bringing its total planned U.S. commitment to $265 billion.

[Link] Mixx Technologies has acquired Bengaluru-based semiconductor IP design firm Sophic Silicon Technologies and announced a manufacturing partnership with Kaynes Semicon to establish an advanced optical integrated circuit test and assembly facility.

[Link] Tower Semiconductor has launched a dual-track expansion in Japan, committing $4 billion to significantly increase its 300 mm silicon photonics, silicon germanium, and advanced packaging capabilities.

Energy

• [Link] Genesis Mission Summit 2026. The inaugural Genesis Mission Summit centered on accelerating AI-driven scientific discovery and high-performance computing through broad collaboration between the government, national laboratories, and industry partners. However, the event highlighted a critical, familiar challenge regarding how to maintain security and ensure absolute algorithmic trustworthiness when deploying AI systems across the nation’s most advanced research infrastructure.

• [Link] Great Sky Selected for DOE’s Genesis Mission. Great Sky’s project, developed in partnership with Brookhaven National Laboratory, will advance highly connected superconducting optoelectronic neuromorphic systems for scalable AI, addressing Genesis Mission Challenge Area 9, Recentering Microelectronics in America. Great Sky’s project addresses a core challenge facing DOE’s scientific mission: processing streaming data from world-class experimental facilities including synchrotrons, particle colliders, and fusion reactors in real time, at a scale conventional AI hardware struggles to support. Note to self…. what the heck are superconducting optoelectronic networks?

• [Link] Congressman Brett Guthrie, Chairman of the House Committee on Energy and Commerce, released a discussion draft of the Permitting Our Way to an Energy Resurgence (POWER) Act, a package of Energy and Commerce Committee bills, to address America’s broken permitting system and lower energy costs for hard-working families.

• [Link] Bypassing the Grid: How Data Center Developers Are Building Their Own Power Plants. Cleanview CEO Michael Thomas performs a thorough analysis of 59 behind-the-meter data centers with ~90 GW of announced capacity.

• [Link] Global electricity demand growth set to accelerate as power systems adjust to recent shocks. The IEA’s latest Electricity Mid-Year Update forecasts that global electricity demand will grow by 3.6% in 2026 and by a further 3.8% in 2027 – up from 3% growth in 2025. Global electricity consumption is projected to reach 30,700 terawatt-hours (TWh) in 2027, compared with 28,600 TWh in 2025.

• [Link] EPRI Current Podcast: How Could Data Center Growth Affect Future Electricity Costs?

The AMD Advancing AI conference was last week, which highlighted new Helios rackscale solutions and much more.

• [Link] AMD launched its next-generation high-performance computing and physical AI portfolios, highlighted by the debut of its first rack-scale AI solution, AMD Helios. Featuring 6th Gen EPYC CPUs and Instinct MI400 Series GPUs, Helios is built for frontier training, inference, and agentic AI workloads while delivering up to 30% more tokens per dollar than competing solutions. Cerebras and AMD are collaborating to deliver a combined solution of Cerebras ultra-low-latency AI compute and AMD Helios high-throughput rack-scale infrastructure to help improve inference efficiency, scalability, and economics for ultra-low-latency inference serving.

• [Link] AMD CEO Lisa Su keynote

• [Link] Data Center Richness: AMD Helios Ultra-Wide Racks Enter the Market

• [Link] TensorWave announced it has adopted the AMD Helios rackscale solution, powered by AMD Instinct MI455X GPUs, to advance its infrastructure for frontier-scale inference and training workloads.

[Link] CuspAI has announced the AI Materials foundry, working with NVIDIA, Meta, Samsung, Hyundai Motor Group, Henkel, Applied Materials, Tokyo Electron and Lam Research among others. The foundry is a global network of data, labs, compute, and scientific expertise for the design of new materials. The company also announced a $450 million Series B led Kleiner Perkins and NEA, with significant participation from Jeff Bezos’ Bezos Expeditions.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

Data Centers

• [Link] SpaceX is in talks ‌to provide the U.S. ‌Department of Defense with access to data center capacity worth billions of dollars to run AI models, the Wall ‌Street Journal ⁠reported on Friday, citing people familiar with the ⁠matter.

• [Link] Elon Musk has quietly acquired APR Energy, a mobile gas and diesel turbine company, in a deal valued at over $1 billion. The acquisition is intended to provide immediate, off-grid power to fuel the high energy demands of xAI’s Grok supercomputers.

• [Link] Meta has announced a massive expansion of its Hyperion AI data center in Richland Parish, Louisiana, increasing its planned computing capacity to 5 gigawatts and raising its total projected investment to over $50 billion.

• [Link] Meta is in talks to lease computing power to Anthropic in a potential deal worth up to $10 billion over two years, according ‌to a source familiar with the matter.

• [Link] Crusoe and Lancium have announced a partnership to build a 1 gigawatt AI data center campus spanning 270 acres in Childress, Texas, with construction set to begin in Q3 2026. Purpose-built for a leading hyperscale technology company, the campus will feature water-efficient liquid cooling and behind-the-meter solar infrastructure while creating over 100 long-term local jobs.

• [Link] QTS and Lancium announce data center campus in Hall County, Texas. QTS Data Centers and Lancium have announced plans to develop a multi-building data center campus in Hall County, Texas, which is expected to bring over $10 billion in capital investment and create thousands of jobs while utilizing independently funded, sustainable energy and water solutions.

• [Link] SK Telecom has announced its plan to build a 15GW AI data center with the goal of establishing South Korea as Asia’s primary AI infrastructure hub. First, starting with the AI data center currently under construction in Ulsan, SKT will build a cluster of over 2GW across the southeastern region (Gyeongsang), using it as a base to attract AI infrastructure demand from global big tech companies to Korea. It also plans to build an additional 1GW in the southwestern region (Jeolla), bringing its domestic AI data center capacity to a total of 5GW to be opened in stages starting in 2029.

• [Link] ACS Group and Global Infrastructure Partners have officially launched Coravel, a global digital infrastructure joint venture designed to simplify the development and operation of next-generation data centers. Alongside the launch, the platform secured its first major hyperscaler customer agreement, covering approximately 140 megawatts of critical IT load across three facilities in the Dallas-Fort Worth area.

• [Link] Rich Miller made a guest appearance on the Shared Everything podcast from VAST Data, hosted by Nicole Hemsoth Prickett. Rich and Nicole explore the significant disconnect between ambitious AI infrastructure goals and the practical constraints of power, advanced liquid cooling, and construction resources required to deliver them.

Quantum

• [Link] Very cool video from Sachin and Adam, where they tour the PsiQuantum factory in Milpitas, California, and talk with co-founder and CSO Pete Shadbolt. Pete explains the building blocks of the machine, and explained why photons, silicon, and standard chip factories might be the path to a million qubits.

• [Link] QuiX Quantum has announced Carina, the world’s first universal photonic quantum computing architecture designed for commercial deployment in customer data center environments. Developed for the German Aerospace Center’s Quantum Computing Initiative (DLR QCI), the room-temperature, compact system serves as a physical qubit foundation to seamlessly integrate with high-performance classical computing and AI infrastructure.

• [Link] Cisco is now testing Aliro Orchestrator as the quantum network management layer at its Cisco Photonics Center in Vimercate, Italy, to operate and manage a live entanglement-based quantum key distribution (eQKD) network built on zerothird’s eQKD system.

Space

• [Link] NASA has selected SpaceX to deliver laser communications capabilities for next year’s Artemis III mission, enabling the agency to provide a front row seat for viewers from the Orion spacecraft as four astronauts test rendezvous and docking capabilities between Orion and test versions of commercial human landing systems needed to return astronauts to the Moon in 2028.

• [Link] The Starlink V3 satellite is designed to support 1 Tbps of downlink and 160 Gbps of uplink capacity. That’s an ~10x improvement in downlink capacity and an ~22x increase in uplink capacity from the Starlink V2 satellite. The V3 satellites phased arrays are fed by upgraded chips that support a ~64x increase in throughput handled per modem chip.

• [Link] Amazon, Globalstar, Iridium, and Telesat have launched the SpaceConnect Association, a new trade organization designed to advocate for the regulatory and market interests of non-geostationary orbit (NGSO) satellite operators.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

Data Centers

• [Link] TeraWulf has signed a 20-year lease agreement with AI company Anthropic to build out a 401 MW purpose-built AI infrastructure campus at its Justified Data site in Hawesville, Kentucky, which is expected to generate approximately $19 billion in contracted revenue. Concurrently, TeraWulf is selling its entire 50.1% majority stake in the 168 MW Abernathy Joint Venture in Texas to an investor group led by Fluidstack, capitalizing on its initial $450 million investment at a premium.

• [Link] Meta announced it is breaking ground on a new 1GW data center in Sturgeon County, Alberta, its first data center in Canada and 33rd in its global fleet. The project represents an investment of over CAD$13 billion, and the company is also investing ~$60 million in local infrastructure improvements.

• [Link]Amazon Web Services has filed permits for “Project Eagle,” a planned $1.2 billion, four-building data center campus on a 2,700-acre site in Wharton County, Texas, southwest of Houston.

• [Link] Cedar Rapids (Iowa) confirms third data center planned for southwest side. Cedar Rapids Mayor Tiffany O’Donnell confirmed Wednesday that a third data center is planned for the city, with the facility set to be built on the southwest side. The confirmation comes as Linn County recently adopted a temporary moratorium and new regulations for large data centers. O’Donnell said Cedar Rapids is not considering a similar moratorium because the project complies with the city’s existing zoning.

• [Link] An unnamed developer has proposed building a “small to mid-sized” data center campus on a 46-acre site in an industrial park in Maquoketa, Iowa. While city officials support the project because the developer is not seeking public incentives and will fund all necessary infrastructure upgrades, the proposal faces notable opposition from local residents.

• [Link] Ryan Shaw YouTube: The Everything Shortage Is Here | Why Tesla Saw It Coming. An excellent video on how the massive AI boom is triggering an “everything shortage”—from memory chips to copper—and why it’s driving up consumer tech and automotive prices. Ryan breaks down the complex intersection of digital demand, physical supply shocks, and geopolitical leverage.

• [Link] SemiAnalysis Weekly: Stop Saying Half of 2026 US Datacenter Capacity Is Canceled

Quantum

• [Link] Core Memory Podcast: Quantum Computers Will Crack Encryption Sooner Than You Think. Awesome interview with Prineha Narang. A quote from her near the end wraps up several of the questions I have had about just attempting to understand quantum computing concepts. “Fault-Tolerant Quantum Computing (FTQC): The guest explains that the threshold for meaningful error correction and utility-scale application is on the horizon. Depending on the physical modality (light, superconductors, neutral atoms, photonics, etc.), some systems could reach this threshold as early as 2028, and others before 2030.”

• [Link] Oratomic has secured $300 million in Series A funding, co-led by ARCH Venture Partners, Spark Capital, Bezos Expeditions, Index Ventures, General Catalyst, Lowercarbon Capital, Bain Capital, Formation, Nebular, David and Scott Aaronson, and others to pursue a challenging goal: building the world’s first fault-tolerant quantum computer. The company is taking a different approach to quantum error correction, utilizing “reconfigurable atomic arrays trapped in focused laser beams” rather than pursuing incremental advances.

• [Link] In a recent interview, Kristen Pudenz, the vice president of research programs at Atom Computing, highlights that rapid advancements in scaling are finally allowing quantum error correction to be implemented at a useful level. This milestone bridges the gap between hardware limitations and utility-scale quantum computing, making fault-tolerant systems a near-term reality rather than a distant goal.

• [Link] QC Design announced the publication of “Plaquette: A hardware-aware design platform for fault-tolerant quantum computers”, the paper presenting the theoretical framework and software suite behind its flagship product. The paper describes how Plaquette computes the logical performance of fault-tolerant architectures directly from the physics of a device’s actual imperfections, and is now available on arXiv.

• [Link] Scientists from Oak Ridge National Laboratory, Cleveland Clinic, and IBM have achieved the first-known instance of computing the molecular configurations of a fusion energy material using a quantum computer. By leveraging quantum-centric supercomputing, the team modeled candidate materials to help optimize the production and extraction of tritium, addressing a critical bottleneck to realizing abundant fusion energy.

Sovereignty. The topic of AI sovereignty has come up quite a bit in my news feed. There were several stories from this past week that I thought I would share here:

• [Link] Toward a Practical Definition of Sovereign AI - Dan Wright, Co-founder and CEO of Armada. True sovereign AI should not be measured by unrealistic standards of complete independence, but rather by an organization’s or nation’s practical ability to deploy trusted existing technology quickly, maintain ownership of its fine-tuned data, build resilient and distributed edge infrastructure, and efficiently leverage available power sources

• [Link] Palantir Technologies announced a strategic initiative with NVIDIA to deliver an intelligent engine for running NVIDIA AI and Nemotron open models in sovereign environments, with a focus on United States government agencies and U.S. critical infrastructure. Government agencies and critical U.S. companies can employ the NVIDIA-Palantir joint offering to retain sovereignty over their AI systems, giving them the ability to control and adapt open models, especially in regulated domains. Palantir also posted on X on the importance of AI sovereignty.

• [Link] Eight NATO allies have announced plans to combine their individual national defense satellites into an interoperable network called the Hybrid Alliance Layered Operations in Space (HALO) constellation. This initiative allows participating nations to maintain control over their sovereign fleets while collaborating to improve high-speed communications, intelligence, and missile tracking capabilities.

• [Link] HUMAIN, a PIF company building full-stack artificial intelligence capabilities globally, and Cohere, the leading sovereign AI company, announced a strategic AI compute collaboration that will provide dedicated AI infrastructure for the next generation of frontier AI models. The companies will also collaborate on the development of enterprise AI solutions and special-purpose sovereign AI models.

• [Link] Germany Expansion: ICEYE formally established a German entity based in Berlin and Munich and appointed Philipp Herkelmann as its country CEO. The new entity will manufacture satellite components, develop advanced software, and create an AI and intelligence center.

• [Link] Portugal Expansion: ICEYE established “ICEYE Portugal” and opened a new center of excellence in Lisbon, naming Rui Costa as the local CEO. The expansion aims to develop advanced space technology, AI, and software in collaboration with Portugal’s existing space ecosystem and the Portuguese Air Force, which has already acquired four ICEYE satellites to monitor its maritime domain.

• [Link] Seven Boson Group announced the launch of its Sovereign AI Decision Intelligence software and services, enabling nations to own, deploy, and govern frontier AI without dependency on foreign providers.

Energy

• [Link] Researchers from Extropic Corp. and MIT have proposed a Denoising Thermodynamic Computer Architecture (DTCA) that could perform AI inference tasks using roughly 10,000 times less energy than conventional GPUs. Although simulated benchmarks demonstrate significant cost and energy efficiency, the full architecture currently remains theoretical, with only the random-number generator circuit physically built so far.

• [Link] Siemens and FuelCell Energy have signed a memorandum of understanding to collaborate on accelerating the deployment of scalable, distributed fuel cell power solutions. Under the agreement, Siemens will design and supply electrical infrastructure to pair with FuelCell Energy’s technologies, targeting rapid commercial deployments for energy-intensive applications like data centers and utilities.

• [Link] In a major milestone for the next generation of nuclear energy, Aalo Atomics announced that its Critical Test Reactor (CTR) successfully achieved first criticality on July 4, 2026. Going from groundbreaking to a sustained nuclear chain reaction in less than eight months, the project represents one of the fastest reactor builds in the last 80 years.

• [Link] The U.S. Department of Energy’s Office of Electricity has released a draft of the 2026 National Transmission Needs Study for a 60-day public comment period ending September 7, 2026. The study highlights an urgent requirement for expanded electric transmission infrastructure driven by rising demand from data centers, expanding manufacturing, and general economic growth. It identifies critical areas of high transmission congestion—primarily within the NYISO, NorthernGrid South, and MISO regions—and underscores the value of interregional links to enhance grid reliability and resilience.

• [Link] The U.S. DOE has closed a loan of up to $3.26 billion to AEP Texas to finance a portfolio of nearly 100 transmission projects, the agency’s Office of Energy Dominance Financing (EDF) said on July 8. The financing will fund the rebuilding, reconductoring, and new construction of roughly 2,800 miles of transmission lines across south and west Texas, AEP said.

[Link] Low Earth Orbit (LEO) satellites require highly precise sensing and positioning to align antennas, cameras, and solar arrays according to specific mission requirements. To achieve this, Attitude and Orbit Control Systems (AOCS) use sensor fusion to combine data from tools like star trackers and IMUs into an accurate estimate of the satellite’s state. The flight software then processes this information to execute real-time orientation corrections using actuators like reaction wheels, magnetorquers, and thrusters.

[Link] Driven by the urgent need for consistent clean energy to power AI data centers and dropping rocket launch costs, space-based solar power (SSP) is transitioning from science fiction to a viable commercial reality. Private companies like Virtus Solis, Overview Energy, and Reach Power are actively developing technologies to collect solar energy in orbit and wirelessly beam it back to Earth, offering a geographically flexible, 24/7 source of baseload power.

[Link] Disclaimer: I am a huge fan of Marc Andreessen…. so this is frigg’n awesome. Federal Reserve Chair Kevin Warsh has enlisted Marc Andreessen — the venture capitalist, adviser to President Donald Trump and one of Silicon Valley’s most aggressive promoters of artificial intelligence — to help shape how the central bank thinks about the technology’s economic impact.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

Data Centers

• Iowa updates:

  • Cedar Rapids passed a data center moratorium. At the Linn County Board of Supervisors meeting last week the Board voted two to one to pass the moratorium, saying that they do not have enough information about data centers to make sound decisions.

  • I am also still tracking this enormous data center in Norwalk (south of Des Moines). The Norwalk City Manager said work is moving forward to prepare approximately 300 acres owned by Tract for future development.

  • A new EEI Electric Perspectives podcast features Alliant Energy and QTS on Cedar Rapids Data Center Campus, Community Benefits, and Customer Affordability. Large-load growth can stabilize rates when structured correctly

• [Link] According to planning documents, domestic developer NextDC has proposed building a 612MW data center named NextDC S7 in Sydney’s western suburb of Eastern Creek on land acquired in October 2024. OpenAI Stargate is planned as a customer.

• [Link] Digital Realty has agreed to acquire Blackstone’s majority equity interest in three fully leased Northern Virginia data centers for $3.5 billion, representing a gross asset value of $7.8 billion.

• [Link] Bloom Energy and Brookfield announced the expansion of their strategic partnership as Brookfield increases its framework to finance power projects for AI infrastructure – from the previously announced $5 billion to $25 billion – a fivefold expansion since October 2025. The increased funding will help grow the fuel cell partnership globally.

• [Link] Meta Unveils ‘Meta Compute’ Cloud Business to Sell Excess AI Infrastructure to Outside Customers. I like this SemiAnalysis take on the news as well. “Capex in 2027 will be shockingly high. In just the first six months of the year, Meta has contracted over 5GW of capacity across Cloud & Colo, and that doesn’t even include all their accelerating self-build activity.”

• [Link] South Korea has announced a $919 billion investment in three tech-focused “mega projects” that include building 18.4GW of AI data center capacity by 2035 to secure global leadership in artificial intelligence.

• [Link] Elea Data Centers and AXIA Energia are partnering to build the Amazon region’s first AI-focused data center in Belém, Brazil, which will launch in 2027 with an initial 7.5 MW capacity powered entirely by 100% renewable energy.

• [Link] DigitalBridge and JEXI announce the formation of Nippon Gateway Infrastructure (NGI), a new Data Center platform in Japan. NGI launches with a foundational portfolio of data center assets acquired from NEC, comprising facilities in the Greater Tokyo and Greater Osaka regions.

• [Link] Data Center Richness: Scaling Data Center Construction for the AI Era. Rich and Doug Mouton discuss the complexities of building at AI scale, and the supply chain and staffing challenges faced by large campus projects.

Energy

• [Link] As the U.S. celebrates 250 years, the U.S. Energy Information Administration reviews energy consumption across the country’s history. Over the past 250 years, the nation’s energy landscape has evolved from early wood-burning into today’s sophisticated mix of nuclear, renewable, and hydrocarbon technologies.

• [Link] The Federal Energy Regulatory Commission (FERC) has issued tailored show-cause orders directing the nation’s six regional grid operators to either justify or reform their rules for connecting large energy users, such as data centers and manufacturing facilities, to the electric grid. Within 60 days, these grid operators must address five specific areas of reform, which include improving application processes, preventing cost shifting, and accommodating co-location agreements. The targeted action aims to modernize electric markets and accelerate the integration of high-demand infrastructure while protecting consumers and ensuring reliable power.

  • [Link] FERC also issued a notice and agenda for the Commission-Led Technical conference on PJM governance and stakeholder reforms.

  • [Link] GridLAB commentary: FERC’s new load flexibility framework is only half the battle. Now, apply it to generation.

• [Link] The UK’s National Energy System Operator (NESO) warns that the cost of modernizing Great Britain’s power grid through the 2030s has risen to £89 billion ($119 billion), marking a roughly 50% increase from previous estimates due to inflation and accelerated clean energy initiatives. This upgraded infrastructure is necessary to connect new low-carbon energy projects and meet a projected 30% surge in electricity demand driven by transport, heating, and data centers.

• [Link] European energy company SGE has submitted an application under the UK’s Advanced Nuclear Framework to build a privately financed £35 billion fleet of 14 GE Vernova Hitachi BWRX-300 small modular reactors across three British sites, aiming to deliver 4.2 GW of clean capacity by 2034.

• [Link] NatPower and Tesla have entered into a multi-year supply and execution agreement covering more than 25 GWh of battery energy storage systems (BESS) across European markets. Projects under the agreement will be sited in Italy and the United Kingdom and will be owned and operated by NatPower. Tesla will provide Megapack, their battery energy storage system, as well as EPC and bankable trading services.

• [Link] National Grid Ventures (NGV) has agreed to invest $1.75 billion for a 35% stake in Joulent LLC, a strategic partnership aimed at developing integrated power solutions and infrastructure for large load customers like AI data centers. The investment will initially fund Project Kilby, a 2.67 GW co-located power facility in West Texas developed in partnership with Chevron to provide dedicated electricity to a Microsoft data center.

Space

• [Link] Rocket Lab acquires Iridium. Rocket Lab has entered into a definitive agreement to acquire satellite communications provider Iridium for approximately $8.0 billion in a cash-and-stock transaction, creating a fully vertically integrated space company that unifies launch capabilities with a global satellite network.

• [Link] Peter Diamandis: A Quadrillion-Dollar Future. PwC projects that the lunar economy will reach roughly $170 billion annually by 2040 through exploration, surface operations, and logistics, serving as a stepping stone toward the even larger prize of utilizing local water and metals for space-based manufacturing and fueling. According to McKinsey and the World Economic Forum, the global space economy is projected to balloon to $1.8 trillion by 2035—almost tripling its 2023 valuation. Even this massive figure understates the explosive, long-term growth expected from emerging sectors like space-based data centers and asteroid mining.

• [Link] SpaceX AI revenue. Following up on the SpaceX compute buildout, Mach 33 now looks at the revenue. Mach33 forecasts SpaceX’s AI and compute segment revenue will grow from $19.5 billion to a base case of $205 billion by 2030.

• [Link] Venture capital firm 021T Capital has invested in Lonestar as the company builds its vision of the first Sovereign Dyson Swarm for AI and Storage infrastructure to serve all users in all nations.

• [Link] SpaceNews: Making satellite communications more resilient. Mike Gruss talks with Kymeta’s Ryan Stevenson. They discuss the benefits and challenges of multi-orbit and multi-band connectivity in a single terminal.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

[Link] Tesla has filed a trademark for “MEGAPOD,” which is described as a self-contained, modular hardware platform designed for artificial intelligence computing. This move aligns with Tesla’s growing focus on AI infrastructure and could support the rapid deployment of distributed computing resources, such as its “Digital Optimus” project. This technology works with Tesla’s other AI projects, like the Dojo supercomputer and self-driving systems. Instead of using traditional, massive data centers, it allows AI tasks to be processed locally or regionally.

[Link] Mach 33: SpaceX’s Compute Supply: The Option to Be a Hyperscaler. SpaceX is built on a highly reliable infrastructure path projected to reach approximately 1.3 million GPUs (3.6 million H100-equivalents) by mid-2029, though the buildout will eventually face power constraints before chip shortages. How the company classifies this massive capacity depends entirely on its own internal usage decisions; retaining less compute shifts it toward a revenue-generating hyperscaler model supported by $28 billion in annual lease contracts, while keeping more for heavy internal inference and training reserves creates an internal cost center.

[Link] SpaceX has signed a computing power agreement with Reflection AI, an Nvidia-backed open-source AI startup, granting it immediate access to Nvidia GB300 chips at the Colossus 2 data center in Memphis, Tennessee, per CNBC reporting published June 22, 2026.

[Link] If I mention SpaceX 3 times, maybe I’ll hit some search engine optimization measure. :) @SawyerMerritt reports that regulatory approval has been received for SpaceX to acquire startup Mesh Optical Technologies, a company founded by former SpaceX engineers working on optical data center communication technology.

[Link] Tom Shaw YouTube channel: Tour of Meta data centers in Altoona, Iowa. It feels like yesterday that I was watching Rich Miller’s tour of the then-new Altoona site. Tom Shaw toured the now-enormous Meta campus in the great state of Iowa and was granted a pretty comprehensive behind-the-scenes look at the infrastructure. I’ve driven around the campus dozens of times (stalker mode)… maybe I should ask for an official tour.

[Link] Infrastructure Research Substack: The $10B campus that shows what Meta is really betting on. Yair Titelboim looks beyond the press release to resolve whether the equity in a deal like Meta’s Indiana data center campus is mostly real estate and execution, or mostly pricing power tied to AI demand that can outrun new supply. Very nice analysis!

[Link] Chevron has signed a 20-year agreement to supply natural-gas-fired electricity to power Microsoft’s massive “Project Kilby” AI data center in West Texas. The dedicated, off-grid facility is expected to consume nearly 2.7 gigawatts of power.

[Link] At the ISC 2026 event last week in Hamburg, Germany, China’s LineShine supercomputer debuted at the number one spot on the Top500 list. LineShine (National Supercomputing Centre in Shenzhen) achieved 2.198 Exaflop/s on High Performance Linpack, about 80 percent of its 2.736 Exaflop/s theoretical peak — making it the first system on the TOP500 to exceed two exaflops of sustained double-precision performance using CPUs only. LineShine draws approximately 42.2 megawatts of power, for an efficiency of 52.07 Gigaflops/Watt.

[Link] The Tech Capital: How Applied Digital secured $31 billion and 1.2GW in just one year. At Datacloud Global Congress in Cannes, Applied Digital founder and CEO Wes Cummins discusses the company’s remarkable growth, the future of AI infrastructure and why small communities are becoming key destinations for large-scale data center investment.

Chips

• [Link] Q.ANT successfully demonstrated a diffusion model and a recurrent neural network on its second-generation Native Processing Unit (NPU) at ISC High Performance 2026 in Hamburg. This proves that Q.ANT’s photonic architecture supports the full breadth of modern AI capabilities in generative image synthesis and sequential time series prediction.

• [Link] Qualcomm and Meta have announced a strategic, multi-generation agreement for Qualcomm to supply data center CPUs for Meta’s server infrastructure. Under the partnership, Meta’s next-generation server fleet will be powered by Qualcomm’s Dragonfly C1000 CPU, which is slated to enter production in the second half of 2028.

• [Link] IBM has introduced the world’s first sub-1 nanometer chip technology, utilizing a revolutionary three-dimensional “nanostack” architecture to deliver up to 50 percent more performance or 70 percent greater energy efficiency than its previous 2 nm node.

• [Link] OpenAI and Broadcom have announced “Jalapeño,” a custom AI inference accelerator built from the ground up for large language models. Architecturally, the chip reduces bottlenecks by balancing compute, memory, and networking resources to minimize data movement and maximize utilization. Early lab testing at production-target frequencies demonstrates a substantial improvement in performance-per-watt over current state-of-the-art hardware when running heavy frontier workloads like GPT-5.3-Codex-Spark.

• [Link] Edge AI on-chip processing and learning company BrainChip has launched its Akida Communication Reference Platform, a physical development tool powered by the AKD1500 neuromorphic processor that enables real-time, on-device RF signal classification. Operating on sub-watt inference power, this platform allows defense contractors and government agencies to prototype portable, battery-powered signal intelligence applications capable of adapting to emerging wireless threats in the field.

Space

• [Link] Rocket Lab announced a definitive agreement to acquire Iridium Communications in a cash-and-stock deal valued at approximately $8 billion. The acquisition merges Rocket Lab's launch and satellite manufacturing capabilities with Iridium's low Earth orbit satellite network, shifting Rocket Lab's revenue model toward stable recurring subscription services to challenge competitors like SpaceX.

• [Link] Last week SpaceX completed a Starfall Demo mission (on the Falcon 9 launch). The purpose of Starfall is to support the “transport and delivery of goods through space.” Most of what is known about Starfall comes from the FAA’s environmental assessment. In that document, the FAA writes that Starfall will “enable point-to-point delivery of critical cargo through space on rapid timelines” and provide access to space for commercial in-space manufacturing, a nascent market that, so far, is largely geared toward pharmaceuticals.

• [Link] Sophia Space announced it has selected Apex, a supplier of productized satellite platforms, to supply a satellite bus to bring real-time orbital computing to enterprise, government, and defense customers. Sophia Space will fly one of Apex’s Nova bus platforms for a demo mission in 2027, demonstrating Sophia Space’s TILE (Thermal Integrated LEO Edge) compute modules.

Energy

• [Link] Data Center Richness: Tapping Smart Homes to Power Data Centers. To address the growing energy demands of AI-driven data centers, companies like Tesla, Sunrun, and Renew Home are collaborating to leverage virtual power plants (VPPs) powered by smart home devices. This initiative aims to aggregate and unlock up to 16 gigawatts of distributed energy capacity to relieve strain on the electrical grids supporting these computing facilities.

• [Link] In Munich, Germany, CATL (Contemporary Amperex Technology Co., Limited)—a Chinese battery manufacturer headquartered in Ningde, China—officially launched the TENER Sodium Energy Storage System, the world’s first field-validated sodium-ion energy storage solution. This new system offers modular flexibility, enhanced safety, and lower operational costs compared to conventional lithium systems, making sodium and lithium the dual foundation for future energy infrastructure

• [Link] This YouTube video looked pretty cool: HYLIION: 3D Printed Power Plant of the Future. Exclusive Hyliion factory tour featuring the groundbreaking KARNO Generator and the future of power generation technology.

• [Link] FuelCell Energy and Fit Energy USA, a developer of reliable power solutions to support advanced computing infrastructure and artificial intelligence, announced a strategic agreement for up to 380 megawatts of clean, baseload on-site power for data centers using FuelCell Energy’s utility-scale fuel cell technology. The agreement includes an immediate deposit for an initial 30 MW of power scheduled to begin delivery later this year.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.

What I watched last week

• [Link] Moonshots episode 265. I absolutely love this podcast. Watch the entire episode, but one particular AWG quote I enjoyed (around 1:14:09) was: “The elephant in the room is terrestrial versus orbital data centers. So if you just extrapolate this trend out, you find that for the foreseeable future, the largest data centers need to remain terrestrial, while perhaps by the early 2030s, barring some left turn in civilization, much of the marginal new compute will remain in orbital data centers. So we have this sort of weird bipolar civilizational compute footprint where if you need to do very large coherent training runs, you’ll do it on land. But if you want inference where you don’t necessarily care about coherence, you’ll deploy it to orbit and sort of squint…. maybe Earth, at least until we figure out how to build very large coherent clusters, not terrestrial. Earth becomes the training hub and space becomes the inference hub. And it’s this sort of weird gradient between them. We build on land, we deploy to orbit. That’s I think the regime we’re likely to find ourselves in, again, barring some left turn in the early 2030s. The question and the elephant in the room is at some point if Elon’s Dyson swarm or competitors’ Dyson swarms come to pass, someone is surely going to look around and ask again barring some radical advance in distributed training. Why do we have so much compute in orbit that we’re not using for training, and why aren’t we building larger clusters?

• [Link] Data Center Richness: What Solid State Transformers Mean for Data Centers. DG Matrix CEO and cofounder Haroon Inam joins to discuss solid-state transformers.

• [Link] Following up from their newsletter article on 800V DC, the SemiAnalysis Weekly podcast interviews Haroon Inam, CEO and Co-Founder of DG Matrix; joined by Jordan Nanos, Jeremie Eliahou Ontiveros, and Nicolas Bontigui to explain why megawatt GPU racks make 800V DC a necessity.

• [Link] Ryan Shaw: Tesla Is The New Data Center King | SpaceX Merger Is Coming

• [Link] Bloomberg: How Data Centers in Space Could Change AI

• [Link] Above the Cloud: Building Data Centers in Space - Richard Campbell - NDC Copenhagen 2026. Richard Campbell explores the challenges and potential of putting data centers into orbit

Data Centers

• [Link] ARK Invest Director of Research, AI, and Cloud Frank Downing discusses the economics of building a gigawatt data center. I really like the data source that Frank uses (Epoch AI): [PUBLIC] US Data Center TCO.

• [Link] Google has announced a $1.5 billion investment for 2026 and 2027 to expand its data center campus in Jackson County, Alabama, fully funding its own power and infrastructure costs. Alongside this expansion, the company is supporting the local community by launching a $2 million Energy Impact Fund for weatherization programs and donating $550,000 for student STEM kits.

• [Link] According to a Bloomberg report, Meta has signed an agreement to secure 1.6GW of compute capacity from Crusoe’s upcoming data center campuses in Childress, Texas, and Warrenton, Missouri.

• [Link] Egypt has granted a license to Hassan Allam Digital Infrastructure and Data Center Solutions, in partnership with A15, to construct a $400 million data center for cloud computing and AI services.

• [Link] $1.2 billion land deal for second data center project in Salem. The 4-3 Glen Brook Group has announced an application for a zoning amendment in Salem Township, Pennsylvania, to clear the way for a 4,000-acre data center campus project with a potential purchase price exceeding $1.2 billion. The planned development would feature a 39-building campus powered by an on-site electrical generation plant, providing a substantial economic opportunity for roughly 194 local landowners.

• [Link] US data center firm Hyperscale Data is planning to deploy humanoid robots at its facility in Michigan. The company has confirmed the beginning of production through its partner of the first 30 OPR-R2 humanoid robots and expects to begin receiving the components and assembling them for deployment at the Company’s Michigan AI data center campus in the third quarter of 2026.

• [Link] Google has introduced Brazos, a rack-mounted, closed-loop liquid-to-air cooling system designed to easily deploy high-density, liquid-cooled AI and HPC equipment inside existing air-cooled data centers. Operating as a self-contained “sidecar” that supports up to a 60 kW thermal load, this plug-and-play architecture enables simple, one-rack-at-a-time installations without requiring costly facility-level retrofits or chilled water loops

• [Link] The record for computing capacity in a single data center has doubled every 7 months. According to Epoch AI, the record for computing capacity within a single data center has doubled every seven months since August 2024, driven by the rapid expansion of frontier AI infrastructure.

• [Link] US awards $500 million to Nvidia-backed SandboxAQ for finding new chipmaking materials. With the $500 million award from the U.S. Department of Commerce, SandboxAQ will aim to come up with new materials where U.S. chip manufacturing either depends on vulnerable foreign supply chains or where the industry faces shortages and bottlenecks.

Quantum Computing

• [Link] Quantum Pulse Ventures has announced the strategic expansion of its composite pulse photonics platform with Quantum Pulse 2.0 (QP2.0), an architecture that processes light directly within the computational fabric rather than relying on electronic translation. Compatible with existing fabrication processes, the platform drastically improves operational fidelity and offers a ten-fold reduction in physical qubit requirements, a four-fold increase in quantum routing speeds, and up to a 90% reduction in manufacturing costs

• [Link] IonQ has announced the launch of Clavis XG Multiplex, a new Quantum Key Distribution (QKD) product designed to scale quantum security across existing metropolitan fiber networks. The solution allows quantum and classical network traffic to coexist on the same fiber strands, eliminating the need for costly network redesigns or dedicated optical infrastructure.

• [Link] IQM Quantum Computers has deployed “Pathfinder,” a 20-qubit Radiance system, at the Oak Ridge National Laboratory (ORNL), marking the Finnish company’s first on-premises installation on U.S. soil. Operating directly within ORNL’s campus, the system is integrated into their high-performance computing environment to help researchers develop tools for a hybrid quantum-classical ecosystem.

• [Link] Atom Computing has secured over $300 million in total funding, including a $100 million Series C round and a $100 million commitment from the U.S. Department of Commerce, to accelerate the development and global deployment of its commercial-scale, fault-tolerant neutral-atom quantum computers.

• [Link] Alice & Bob has unveiled its first full-stack hardware platform, the Helium Quantum System, transitioning from a chipmaker to a complete system developer. The on-premise system uses 18 physical cat-qubits to encode its first logical qubit, and it is open to research partners to advance fault-tolerant quantum computing and quantum error correction.

[Link] In a recent newsletter, Alex Wissner-Gross dropped the news that Ornn is launching its Ornn Token Price Indices (OTPI). I’ve been tracking this ever since he first mentioned it - and I love the idea of this volume-weighted unit-value index that tracks the real-time, historical transacted price of AI inference tokens. OTPI solves this by tracking the actual transacted market price of AI tokens, pulling daily volume data from real, paid transactions across major frontier labs like OpenAI and Anthropic. The result? A standardized dollar-per-million-tokens benchmark. Ornn’s existing compute index (OCPI) measures the input, or the cost of raw GPU time. OTPI measures the output, or the actual tokens that compute spits out. By pairing the two, Ornn is giving investors and enterprises a clear, dual-sided view of the AI cost curve. It’s a massive step toward tracking real-world demand and the true deflationary curve of machine intelligence.

Energy

• [Link] Critical Energy, a startup founded by ex-SpaceX engineer Spencer Jackson, has secured $22 million in early funding to mass-produce factory-built, modular geothermal turbines. Rather than constructing custom power plants on-site over several years, the company aims to ship container-sized units that can be deployed in weeks to meet the skyrocketing clean energy demands of data centers.

• [Link] Helion has become the first company in the world to secure the necessary regulatory licenses for fusion power plant operations after receiving radioactive materials and air emissions licenses from the Washington Department of Health for its Orion facility.

Space

• [Link] Japan has developed a washing-machine-sized satellite called OHISAMA, which is scheduled to launch in fiscal year 2026 to test space-based solar power by converting sunlight into microwaves and beaming it back to Earth. The primary goal of the mission is to demonstrate highly precise retrodirective beam targeting by successfully powering a single LED on the ground, paving the way for massive 1-gigawatt orbital power stations in the 2040s.

• [Link] SpaceX AI1 (its first orbital compute satellite): A Subsystem-Level Reconstruction. I don’t have access to the article, but I imagine it is awesome.

• [Link] SpaceX Starfall is a mass-producible, uncrewed reentry vehicle, marking its strategic entry into the dedicated in-space logistics and cargo return market. SpaceX is targeting Tuesday (June 23, 2026) for the inaugural launch of its Starfall Demo mission aboard a Falcon 9 rocket from Cape Canaveral. This initial flight serves as a major technology demonstration and operational test for a new class of mass-producible reentry vehicles.

• [Link] LA TIMES - Orbital Data Centers: Coming to a Planet Near You. An increasing number of aerospace companies and tech giants are developing data centers in lower Earth orbit to meet the massive compute demands of AI without adding to the environmental and energy strain on terrestrial power grids. Startups like Sophia Space are leveraging innovative, lightweight, passive cooling technologies to overcome the harsh thermal environment and payload weight limitations of space. Experts project that as launch costs decline and orbital data transfer speeds increase, space-based data processing will become a standard way to enable real-time, edge computing for earth observation and national defense.

[Link] Amazon is in talks to sell its custom Trainium AI chips directly to third-party data centers, mimicking a recent strategy by Google to challenge Nvidia’s dominance amid surging demand for AI hardware. Though Nvidia still holds a massive ecosystem advantage, Amazon’s move capitalizes on its $20 billion in-house semiconductor business and early partnerships with companies like Anthropic and Uber.

[Link] The end of the big tech cash cow: Why the A.I. Infrastructure Buildout Is Destroying the Core FCF Investment Thesis, and Will Fuel a Historic Rotation in Search of Free Cash Flow Machines! The unprecedented AI CapEx boom is fundamentally breaking the asset-light, free-cash-flow model of Big Tech, forcing these former cash cows to dilute shareholders and take on heavy debt. As a result, savvy macro investors are shifting capital into the heavily discounted precious metals and mining sectors, which are quietly generating record-breaking free cash flow margins.

Subscribe now

Follow me on X for more links and news.

Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.