During yesterday’s earnings conference call, CEO Ford Tamer made a statement that immediately caught my attention:

“We also expect revenue to continue to grow at over 30% CAGR for the next five years.”

That’s a remarkable forecast for an established semiconductor company. Tamer was talking about growing revenue by more than 30% annually, compounded, for five years.

And the near-term outlook is even stronger. Management expects revenue to grow at least 60% in 2027, following approximately 90% growth this year.

Those are numbers I normally associate with much smaller, emerging technology companies. Lattice has already become a meaningful supplier of programmable chips used in AI servers, communications equipment, industrial automation, and other applications.

The more interesting part of this story is that the stock has been crushed over the past several months, despite the tremendous improvement in its business.

LSCC traded as high as $157.01 on June 22. It closed yesterday at just $95.44, a decline of approximately 39%. Even after today’s impressive earnings reaction, with shares trading around $107, the stock remains approximately 32% below its June high.

I’ve been averaging into LSCC throughout today’s session, even though the general market is still very dicey, and my current cost basis is approximately $107 to $108 per share.

Here’s why.

An Exceptional Earnings Report

Lattice reported third-quarter revenue of $291 million, an increase of 118% from the same quarter last year. Wall Street had expected approximately $256 million, so the company beat revenue expectations by nearly 14%.

Adjusted earnings came in at $0.72 per share versus expectations of roughly $0.55. That’s an earnings beat of approximately 31%.

Adjusted EBITDA reached $131 million, well above expectations of approximately $108 million.

These are exceptional results, but I think the forward guidance was even more impressive.

For the fourth quarter, management expects revenue between $325 million and $345 million. The midpoint of $335 million is approximately 14% above Wall Street’s previous expectations and represents 130% year-over-year revenue growth.

Management also expects adjusted fourth-quarter earnings between $0.82 and $0.86 per share, compared with prior expectations around $0.65.

In other words, Lattice told investors that the business is accelerating.

The company expects to finish 2026 with approximately $1 billion in revenue and adjusted earnings of roughly $2.50 per share.

For a company that generated $523 million in revenue in 2025, that’s an extraordinary change in the trajectory of the business.

The AI Infrastructure Opportunity

Lattice makes field-programmable gate arrays, or FPGAs. Unlike traditional processors designed to perform a fixed set of functions, FPGAs can be configured to perform specific tasks.

They’re increasingly important in AI infrastructure because servers require chips for security, power management, connectivity, system control, and various other functions.

Lattice isn’t competing directly with Nvidia to build the fastest AI accelerator. It’s supplying chips that help the surrounding infrastructure operate.

And as AI servers become more complicated, the opportunity to sell additional chips into each system increases.

That is exactly what management described during yesterday’s conference call.

Lattice’s Compute and Communications segment generated $171.9 million in third-quarter revenue, increasing 132% year over year. The company’s core FPGA business grew approximately 90%.

That last number is important because Lattice also recently acquired AMI, which contributed $39.1 million during the quarter. The acquisition helped drive the reported 118% consolidated growth rate, but the underlying FPGA business is growing extraordinarily fast on its own.

Management expects global server shipments to increase from approximately 20 million units this year to 25 million in 2027. It also expects AI servers to represent approximately 40% of those shipments, up from approximately 25% in 2026.

Lattice stands to benefit from both the growth in the number of AI servers and the increasing number of programmable components required in each server.

Management Has Visibility Into 2028

One of the things I pay close attention to during earnings calls is whether management can explain why it expects growth to continue.

Anyone can issue an optimistic forecast. I want to know what supports it.

During yesterday’s Q&A, Tamer told analysts:

“We have backlog well into 2028.”

He explained that the company already has orders extending into the first half of 2028, along with multiyear customer supply agreements.

That doesn’t eliminate the possibility of cancellations or changes in customer demand, but it does provide considerably more visibility than a typical semiconductor company might have.

Tamer also discussed the possibility that 2027 revenue growth could reach 70% or even 80%. He was careful not to commit to those numbers, emphasizing that management was comfortable forecasting at least 60% this early.

I interpreted that exchange as evidence of considerable confidence in the business, although I’m using the official 60% outlook when evaluating the stock.

I also liked management’s comments about the industrial business.

Lattice’s Industrial and Embedded segment grew 35% year over year, reaching $80 million in revenue. Management expects that business to continue recovering as customer inventory levels normalize.

So, the company isn’t entirely dependent on AI data center spending. Factory automation, robotics, machine vision, aerospace, and other industrial applications provide additional opportunities.

The AMI acquisition adds another layer through firmware management, security, and control solutions. Management discussed security retrofit products that are entering production now, with potentially meaningful revenue contributions in 2028.

These opportunities help explain why management believes revenue can continue compounding at more than 30% annually for five years.

The Financial Model Is Getting Better

Growth alone doesn’t make a stock attractive. I want to see earnings grow faster than revenue as the business scales.

Lattice appears to be doing exactly that.

Third-quarter adjusted gross margin was 70.9%, up from 69.5% a year earlier. Adjusted operating margin reached 41.1%, while adjusted EBITDA margin was 45%.

Those are excellent margins for a semiconductor company.

Adjusted earnings grew approximately 157% year over year, outpacing the company’s 118% revenue growth.

Lattice also generated $77.6 million in free cash flow during the quarter and paid down $50 million of debt. Management expects net leverage to fall below two times EBITDA by the end of 2026.

There is one accounting distinction worth mentioning. GAAP earnings were just $0.08 per share, compared with adjusted earnings of $0.72. Acquisition-related expenses and other adjustments explain much of that difference.

I wouldn’t ignore the gap between GAAP and adjusted earnings. Over time, I want to see strong growth translate into substantially higher reported earnings and free cash flow.

Still, the underlying operating leverage is difficult to dismiss.

The Valuation Is Where This Gets Interesting

This is the main reason I’ve started accumulating shares.

At approximately $107.50, LSCC has a market capitalization around $15.3 billion.

Using trailing revenue of approximately $809 million, the stock trades at about 18.8 times sales. That’s expensive by conventional standards.

But trailing sales are becoming less useful when a company’s revenue is growing this quickly.

Management expects approximately $1 billion in revenue for 2026 and at least $1.6 billion in 2027. Against that 2027 revenue outlook, the price-to-sales multiple drops to approximately 9.5.

The earnings valuation is more interesting.

Lattice expects approximately $2.50 in adjusted EPS for 2026. That puts the stock at roughly 43 times this year’s expected earnings.

However, analysts are rapidly increasing their forecasts following the earnings release. Rosenblatt raised its 2027 adjusted EPS estimate from $3.20 to $4.58, while its 2028 estimate increased from $4.21 to $6.15.

As of this afternoon, Finviz was showing a next-year earnings estimate around $4.39 per share. The broader analyst consensus is still being revised following last night’s report.

Using $4.39 in 2027 earnings, LSCC trades at approximately 24.5 times next year’s estimated earnings.

If Rosenblatt’s more optimistic $4.58 estimate proves accurate, that multiple falls to approximately 23.5.

For a company forecasting at least 60% revenue growth next year, with adjusted earnings growing even faster, those multiples strike me as attractive.

Consider what the stock might be worth at different earnings multiples using the $4.39 estimate:

2027 P/E multiple

Implied stock price

20x —> $88

25x —> $110

30x —> $132

35x —> $154

These are valuation scenarios, not price targets.

I don’t think a 30-times earnings multiple would be unreasonable for a semiconductor business growing revenue at 60% and potentially sustaining 30% annual growth over the following several years.

That would imply a stock price around $132, approximately 23% above my current cost basis.

At 35 times earnings, the implied value approaches $154, roughly where the stock traded in June.

And that assumes 2027 earnings estimates don’t increase further.

For perspective, AMD and Astera Labs currently trade at forward earnings multiples above 40, while Monolithic Power Systems and Marvell trade closer to 30. Lattice’s multiple is competitive with that group, particularly considering its projected growth rate.

LSCC also screens with a PEG ratio around 0.3 to 0.4, depending on the data provider. That is unusually low for a high-growth semiconductor stock, although PEG ratios can be misleading when earnings are expected to accelerate sharply for a relatively short period.

I wouldn’t buy the stock solely because of a low PEG ratio. I find the combination of earnings growth, operating leverage, and forward valuation much more convincing.

Wall Street Is Starting to Reprice the Stock

Several analysts raised their price targets this morning.

Rosenblatt increased its target from $175 to $200. Baird raised its target from $174 to $195, while Stifel moved from $165 to $175.

These were primarily price-target increases rather than changes to existing Buy ratings.

What matters more to me is what’s happening to the underlying earnings estimates.

Rosenblatt now forecasts $1.63 billion in revenue for 2027 and $2.03 billion for 2028. Its higher price target reflects expectations that Lattice’s growth can extend well beyond the next few quarters.

I’m not buying the stock because an analyst has a $200 price target. Those targets are only as good as the assumptions behind them.

But the magnitude of the earnings revisions tells me Wall Street is reassessing this company’s future earning power.

Why Has the Stock Fallen So Much?

This is the other part of the opportunity.

LSCC reached $157 in June before falling nearly 40% into yesterday’s close.

Part of that decline reflects the broader semiconductor market’s sensitivity to AI infrastructure spending, valuation, interest rates, and concerns about whether hyperscalers can sustain their enormous capital expenditure budgets.

High-multiple semiconductor stocks can experience substantial corrections even when their underlying businesses are performing well.

Some of that concern is legitimate. Growth expectations across the AI semiconductor industry have become enormous, and investors are increasingly sensitive to anything suggesting a slowdown.

But I find it difficult to reconcile a stock trading more than 30% below its highs with a company that just reported 90% growth in its core FPGA business and guided to at least 60% total revenue growth next year.

The earnings report doesn’t guarantee the shares will return to $157, and the previous high isn’t necessarily fair value.

Still, it creates an interesting situation. The stock price has corrected substantially while the company’s future earnings expectations have improved.

That is the type of disconnect I look for.

What Could Go Wrong?

There are several legitimate risks.

First, a meaningful slowdown in AI infrastructure spending would affect Lattice, regardless of its growing market share or strong backlog.

Second, the five-year growth forecast is ambitious. A 30% compound annual growth rate means revenue would grow to more than 3.7 times its starting level over five years. Management has a great deal to execute.

Third, acquisitions can complicate the financial picture. AMI expands Lattice’s opportunity, but integration costs, additional debt, and the difference between adjusted and GAAP profitability deserve continued attention.

Fourth, the stock is still not cheap on conventional trailing valuation measures. At nearly 19 times trailing sales, investors are paying for substantial growth. If earnings estimates decline, that valuation could become difficult to defend.

Finally, the chart still needs work. LSCC is attempting to recover from a significant correction and faces overhead resistance around its 50-day and 200-day moving averages, currently in the $115 to $116 area.

I would like to see the stock establish support following today’s earnings move and then begin working back above those levels.

For now, I’m comfortable averaging into the position because I think the longer-term earnings opportunity outweighs the near-term technical uncertainty.

My Take

What attracted me to Lattice is not simply that the company reported a great quarter.

It’s that management is describing a business that could grow much faster, for much longer, than investors previously expected.

Revenue grew 118% last quarter, including the AMI acquisition. The core FPGA business grew approximately 90%. Management expects approximately 130% revenue growth in the fourth quarter, at least 60% growth in 2027, and more than 30% annualized revenue growth over the next five years.

At the same time, the stock remains approximately 32% below its June high.

And using updated 2027 earnings estimates, shares are trading at roughly 24 to 25 times next year’s adjusted earnings.

I think that’s an attractive combination of growth and valuation, provided management can deliver on its forecasts.

I’m not assuming Lattice will reach Rosenblatt’s $200 target, nor am I suggesting the stock can’t pull back after today’s rally.

But I believe there’s a credible path toward significantly higher earnings over the next several years, and the market may still be underestimating that potential.

I’ve been averaging into LSCC today, with my current basis around $107 to $108.

I’ll be watching the company’s bookings, FPGA content per server, industrial growth, and future earnings revisions closely. Those will tell me whether this is the beginning of a sustained rerating or simply a sharp earnings bounce.

For now, I like what I’m seeing, and here’s a bonus! For a stock that gets very little option activity, I saw these two transactions today:

Disclosure: I own shares of LSCC with a current average cost basis of approximately $107 to $108. My views may change as new information becomes available. This article reflects my personal investment analysis and is not a recommendation to buy or sell securities.

Sources

• Lattice Semiconductor Q3 2026 earnings release

• November 2, 2026 earnings conference call transcript

• Rosenblatt’s revised earnings forecasts and price target

• LSCC stock valuation and earnings estimates, Finviz

• LSCC historical stock prices

And, I have a feeling new traders completely misunderstand this concept. They hear about money managers running a portfolio, and they hear talking heads on CNBC. These professionals are ALWAYS in the market. This is because they frequently have fresh money coming in under management from new or existing investors, and/or, because they compete against an index. Thus, many managers will average down into existing positions with incoming money, and they will also feel good about themselves when the S&P500 is up 3.5% YTD, and the fund they manage is up 4.5% YTD.

Guess what? As a retail trader, you should NOT behave like them. In fact, it’s detrimental, in my opinion, if you are always in the market. If you are an ultra-long-term investor, this is another matter, but this substack is about trading.

So, back to the question I was asked:

Hi sp3cul8r can you give some words of wisdom on how to develop patience in the market and avoiding trades when not favorable? I feel like I always need to be in the market.

I answered with the following response:

1) Develop a feel for price action in the stock market.  Are stocks following through on moves?  If price action is choppy as it has been since June 1, consider staying out or, at least, keeping positions much smaller.

2) Create a rule and stick to it.  I use the following:  I'm aggressively long when SPY is trading really healthy over the 20 day moving average.  When SPY is below the 20 but above the 50 I'm less aggressive. When SPY is below the 50, I'm staying in cash or taking really quick day trades.  When SPY is below the 200, I’m definitely ALL CASH

3) Psychologically, the ability to be out of the market, and in cash, is very important to me.  Trading takes a ton of mental capital. When I feel my brain getting overwhelmed, I will also go to cash.  I’m a big proponent of managing your mental capital this way. If you are exhausted from, say, 2 months of constant trading, you will eventually make a mistake. Also, when you take a break from trading and your mind is refreshed, you see opportunities much better.

4) Finally, your PNL (Profit and Loss) is a signal. For example, if you take three or more losses in a row, market conditions have likely changed. These are your trades giving you feedback that the market has changed and may not be compatible with how much you are trying to take from it. The results of your trades are a very important signal.

5) And here is a 5th point I’m adding for this article. Learn to identify reversals. Hammer candles, bottoming patterns, volume spikes on sell offs - all of these can help you identify when bad market conditions are transitioning to good market conditions.

Here’s a chart of SPY to get you thinking about how market conditions change in the various boxes I’ve drawn. Blue boxes are good market conditions, but the third blue box displayed an immense amount of POWER. That is an environment where it pays to be very aggressive. Conversely, the second red box is basically what we’ve experienced since June 1. While there have been strong reversal up periods within this box, it has been a really tough period to make money.

Bottom Line

As a retail trader, you have limited capital and you are not competing with an index. Your job as a trader is to put your money to work aggressively when conditions are great, and to protect this hard-earned capital when conditions are tough. The market will give you, during a given year, a handful of exceptional time periods in which to make money. Learn to identify those periods, and, learn to stay away or limit your exposure to tough periods. In other words, adapt your trading activity to the market!

Before the war started, OPEC controlled 48% of global oil production. Now, after Iran shot missiles all over the Gulf, taking out oil infrastructure, pipelines and refineries, OPEC now controls under 40% of global production.
As this concept of OPEC’s waning influence dawned on me today, at the same time, something pretty amazing happened in energy markets, also today!

Venezuela's National Assembly approved a massive U.S. oil-development deal that could pull tens of billions of barrels back into a U.S.-aligned supply chain. As a side note, I think Chevron looks like one of the clearest winners here, with its Venezuelan production already around 280k bpd. Management has said, previous to today’s deal between the two countries, that production would rise 50% by 2028 to 420k bpd. This may rise even further tomorrow when the full extent of the deal is revealed. U.S. Energy Secretary Wright is in Venezuela right now and the deal is expected to be announced on Wednesday September 2nd.

So, let me back up a second. I have been asking myself the question, did the U.S. know, before the attack, that Iran would lash out at its neighbors and destroy oil infrastructure thereby weakening OPEC? The answer is “yes”. Two days before the attack on February 28th, CSIS published an analysis asking “How Would Iran Respond to a U.S. Attack?” It specifically pointed to Iran's history of using missiles, drones and proxies against regional targets, including the 2019 attack on Saudi oil facilities. And this wasn't some obscure theoretical possibility. For years, strategic studies had identified Saudi oil-processing facilities, export terminals, power infrastructure and desalination facilities as obvious Iranian retaliation targets in a major U.S.-Iran confrontation. A detailed CSIS study, for example, specifically identified facilities such as Abqaiq as particularly vulnerable.

So, the White House and the National Security Council unquestionably would have modeled the possibility that attacking Iran could cause Iran to attack Gulf oil infrastructure and/or close Hormuz. But, I still find it highly unlikely “oil dominance” was part of the initial thinking. I do believe the initial reasons were to prevent the country from getting a nuclear weapon and to degrade Iran’s military capabilities, especially their arsenal of ballistic missiles. I’m not going to give any President credit for willingly going through the following short-term events to get to LONG TERM energy dominance:

1. 
   1) Oil price spikes including Brent Crude > $100
   2) SPR drawn down to irresponsible levels
   3) U.S. gasoline prices spike - Presidents generally HATE this as it kills them in the polls
   4) Qatar’s LNG exports collapse
   5) Saudi/Kuwaiti infrastructure is damage

2. 6) Hormuz becomes unreliable

If your long-term plan is cheap and secure energy, this sure is a very dangerous way to start such a plan.

HOWEVER, I will say the war probably wasn’t started to destroy OPEC, but once the war dramatically weakened OPEC’s position, the administration realized it had an opportunity to accelerate a Western Hemisphere energy architecture it was already building.

Lets review the timeline:
January
→ Maduro removed
→ U.S. starts taking control of Venezuelan oil flows
→ China/Russia/Iran squeezed out
→ Venezuelan oil sector opened to Western capital.

February 12
→ Wright visits Venezuela
→ talks about dramatic production growth
→ Chevron expansion already underway.

February 28
→ Iran war begins.

March–August
→ Hormuz disruption
→ Gulf production disrupted
→ OPEC+ global oil share plunges from >48% toward 40%.

August/September
→ NABEP and the United States gets 65B barrels of Venezuelan Oil
→ U.S. gets 35% ownership
→ U.S. gets 20% of production at cost
→ U.S. gets first refusal on the other 80%
→ Chevron expands
→ Venezuela considers potentially leaving OPEC

And today, U.S. Energy Secretary Wright took it another step: he said Venezuela's oil production should more than double over the next few years and specifically said the additional production should put downward pressure on oil prices.

This is beginning to look like an energy doctrine to me, that could look like this….

And in 2030, this North/South Americas bloc could be looking at production more like this, which is absolutely OPEC scale or beyond.

So, Trump entered the Iran war already pursuing a strategy to shift global energy power toward the Western Hemisphere. The war unexpectedly, or perhaps foreseeably, crippled a large portion of OPEC's ability to influence the market. The administration now appears to be exploiting that opening aggressively by rebuilding Venezuela as a U.S.-aligned producer.

And there’s one additional twist worth watching closely:

Venezuela is a founding member of OPEC.

If Washington rebuilds Venezuela from ~1.2 mbd toward 2–3 mbd, while encouraging maximum production rather than quota discipline, and Venezuela ultimately leaves OPEC, then Trump’s Venezuela strategy doesn’t merely provide alternative supply.

It takes barrels and potentially a major member away from OPEC itself.

Combine that with the UAE already leaving OPEC, rising Guyana/Brazil/Argentina production, Canadian supply and ~14 mbd of U.S. production, and you begin to see something much larger:

Not the intentional destruction of OPEC through war, but perhaps the intentional construction of a post-OPEC energy system once the opportunity presented itself.

If you’re interested in more in-depth analysis and interaction throughout the trading day, join me in the Tribeca Trading Group Room! Click here and use code SP3 to get 50% off your first month. We look forward to seeing you!

You can also follow me on X here

Subscribe now

I’ve moved my trading accounts to cash.

I’m not bearish and I do not think the AI trade is over. I’m also NOT trying to call a market top.

I simply think the risk/reward of carrying a large book into this week’s two biggest events isn’t particularly attractive.

Nvidia reports Wednesday. Kevin Warsh speaks at Jackson Hole Friday.

I would rather see the cards before placing the next bet.

Nvidia: The Setup Is Getting Interesting

Nvidia reports Wednesday after the close, and it is difficult to overstate how important this earnings report has become.

Nvidia is no longer just another semiconductor company reporting earnings. It has become a real-time barometer for AI spending, hyperscaler CapEx, data-center construction and investor confidence in the entire AI buildout.

But here’s what makes this report especially interesting:

Nvidia has fallen following each of its last four earnings reports. (Investopedia)

Think about that.

The company has continued putting up extraordinary growth, but simply beating expectations hasn’t been enough. The expectations embedded in Nvidia have often been even larger than the numbers Nvidia ultimately reported.

This time, however, something feels different to me.

Nvidia is expected to report roughly $92 billion in quarterly revenue, yet the stock is still about 8% below its May high and has gained only around 16% this year. (Investopedia)

And positioning is surprisingly light.

Morgan Stanley recently identified Nvidia as the most under-owned large-cap technology stock in its institutional ownership work. The gap between Nvidia’s weight in the S&P 500 and its average weighting in actively managed portfolios is unusually large. (Investopedia)

Cantor Fitzgerald made a similar point Monday morning. The firm reiterated its Overweight rating and $350 target, saying investors remain underweight Nvidia and that it expects the stock to move rapidly once it begins moving higher. (StreetInsider.com)

That’s the dislocation I’m interested in.

The fundamentals remain enormous. The stock hasn’t done much. And many institutions apparently aren’t positioned for a major upside move.

That doesn’t guarantee Nvidia goes higher Wednesday. Quite the opposite actually. Another great quarter that gets sold would be meaningful information.

But if Nvidia delivers the kind of beat-and-raise that suddenly causes investors to reconsider their positioning, there could be a lot of money trying to get through the same door at once.

I don’t need to predict which outcome occurs. I can wait and react.

Then Comes Warsh

Two days later we get the other major event: Federal Reserve Chairman Kevin Warsh’s Jackson Hole speech Friday morning.

The Fed currently lists his keynote remarks for 10:00 a.m. ET Friday. (Federal Reserve)

Long-term Treasury yields have recently become one of the biggest pressure points for technology stocks.

Higher yields change the discount rate investors apply to future earnings. And nowhere does that matter more than in long-duration growth stocks and the AI complex. So one comment from Warsh can move bonds, bonds move yields, and yields can quickly reprice equities.

Again, why guess?

I’ve Earned the Right to Wait

Here is where my combined trading accounts currently stand:

August: +11.1%

2026 YTD: +26.3%

One of the lessons I’ve learned in trading is that you don’t have to participate in every market event.

There are moments when having conviction matters and there are also moments when protecting what you’ve already made matters more.

After an 11% month, I have absolutely no need to gamble on whether Jensen Huang says exactly the right words Wednesday night or whether Kevin Warsh strikes the right tone Friday morning.

Cash gives me optionality.

If the market presents a great setup afterward, I can act.

My Best Trades of August

I went back through all three trading accounts and grouped individual fills belonging to the same round-trip position.

Rank

Trade

Realized P&L

1 GDXU +$15,674.63

2 SNDK +$11,428.19

3 SPCX +$8,058.89

4 NVDL +$7,437.09

5 TEM +$6,036.13

Top five winners: +$48,634.93

The SPCX trade was particularly interesting. I accumulated 1,075 shares, including 975 shares at $113.77, took a portion off quickly, and ultimately closed the balance at $137.57. Net profit on the complete position was approximately $8,059.

TEM was another good one. I bought 1,000 shares around $59.69 and sold them the following day in pieces between roughly $65.45 and $66.06, producing approximately $6,036. I still love this stock as part of the Merck/Moderna cancer platform, but it may need to cool off a bit.

My Five Worst Trades

Rank

Trade

Realized P&L

1 AMZU -$3,492.15

2 SPCH -$1,991.24

3 NVDL -$1,881.76

4 PTRN -$1,865.31

5 DELL -$1,731.94

Five largest losers: -$10,962.40

That may be my favorite statistic from the entire month.

My five biggest winning trades generated $48,635.

My five biggest losing trades cost me $10,962.

That’s roughly $4.44 of profit from the biggest winners for every $1 lost on the biggest losers.

That’s trading. I don’t need to be right constantly. I need to keep mistakes manageable and make substantially more money when I’m right than I lose when I’m wrong.

What Happens Next

My checklist from here is pretty simple.

First, let Nvidia report. Don’t just look at whether the company beats estimates. Watch the reaction. Does Nvidia gap higher and hold it? Do semiconductors participate?

If Nvidia puts up another monster quarter and still gets sold, I will likely stay in cash.

Then I want to hear Warsh Friday and watch the reaction in the Treasury market.

If Nvidia validates the AI trade, the stock holds its gains and yields cooperate, I’ll have plenty of opportunities to put money back to work.

If Nvidia gets rejected again or Jackson Hole sends yields materially higher, I’ll be very happy I waited.

There will always be another trade.

For now, cash is my position.

The headline was simple: their personalized mRNA melanoma therapy, intismeran autogene (V940/mRNA-4157), produced the first positive Phase 3 result for an individualized neoantigen therapy. But the bigger story is about a new treatment platform — one that could eventually spread across many types of cancer.

And if that happens, the ripple effects could be meaningful not only for Merck and Moderna, but also for Personalis and Tempus.

This therapy starts with the patient’s own tumor. Each treatment requires sequencing, analytics, target selection, and custom manufacturing. In other words, this is a data and workflow story.

That is where Personalis — and now Tempus, through its planned acquisition of Personalis — enters the picture.

The simplest way to think about who does what

Here is the division of labor:

• Merck brings KEYTRUDA, oncology development muscle, regulatory experience, and global commercialization capability.

• Moderna brings the mRNA platform, neoantigen-selection algorithms, and the ability to design and manufacture a personalized mRNA therapy for each patient.

• Personalis provides the tumor profiling and sequencing layer that helps characterize each patient’s cancer.

• Tempus is acquiring Personalis, which means it could end up owning a valuable piece of the genomic workflow that sits upstream of the therapy.

The vaccine generation process looks like this:

1. A patient’s tumor is removed after surgery.

2. The tumor is sequenced and profiled to identify the mutations that make that cancer unique.

3. Moderna’s system selects the most promising neoantigens from that mutational fingerprint.

4. A custom mRNA therapy is manufactured for that patient.

5. The patient receives the therapy along with KEYTRUDA.

The goal is to train the immune system to recognize the tumor’s unique signals and attack any remaining cancer cells before the disease returns.

That is why this result feels bigger than a typical drug success. The entire approach depends on repeating the same personalized workflow over and over again, one patient at a time.

Why this is a platform, not just one melanoma product

Investors often hear the word “platform” used too loosely. In this case, I think it fits.

A platform is not just a single therapy. It is a repeatable engine that can be applied across many patients and, potentially, many tumor types.

That is exactly what Merck and Moderna are trying to build.

The core logic is the same every time:

• sequence the tumor,

• identify the relevant mutations,

• choose the neoantigens most likely to create an immune response,

• manufacture a personalized therapy,

• give it in a setting where the immune system has a chance to prevent recurrence.

If that process works in melanoma, there is no reason to assume the opportunity stops there. Merck and Moderna already have a broader development program spanning other cancers, including non-small cell lung cancer, bladder cancer, and renal cell carcinoma.

That is what makes this potentially so important.

Melanoma is the first proof point. It may not be the end market.

If the same platform can move from melanoma into larger cancers, then every step in the workflow becomes more valuable — especially the step that has to happen for every patient before the drug can even be built.

That is the hidden appeal of Personalis.

Why Personalis matters more than many investors realize

Personalis is not the company making the drug. It does not own the blockbuster economics the way Merck and Moderna do.

But that does not mean its role is small.

Personalis has said that Merck and Moderna have used its tumor-profiling platform since the inception of the V940 program. It also extended its relationship with Moderna with language suggesting the collaboration could continue into routine patient treatment if the therapy reaches commercialization.

That is a big deal.

It suggests Personalis is not just a background lab vendor that happened to touch the program years ago. It suggests the company is part of the workflow architecture behind one of the most important personalized cancer programs in development.

And now Tempus has agreed to acquire Personalis.

That changes the way I think about the deal.

Before this melanoma result, the cleanest strategic logic behind the acquisition was that Tempus wanted deeper capabilities in MRD and tumor-informed oncology testing. That still matters. But this Merck/Moderna development adds another layer:

Tempus may be buying into the sequencing and analytics layer of a future personalized cancer-vaccine platform.

This could mean Tempus owns a picks-and-shovels position in a therapy category that may scale.

What could Tempus and Personalis actually be worth if this works?

Do not confuse drug revenue with testing revenue.

Merck and Moderna would own the main drug economics. Personalis, and eventually Tempus, would be much more likely to capture revenue from sequencing, profiling, analytics, and workflow support.

Still, that can add up fast if the therapy becomes widely used.

Let’s use a simple illustrative framework.

Assume:

• the eventual net revenue per patient for the drug is roughly $125,000,

• Tempus/Personalis captures about $10,000 per treated patient in sequencing/analytics-related revenue,

• and Personalis remains embedded in the workflow for most patients.

Under those assumptions:

• A $3 billion franchise would imply about 24,000 patients and roughly $216 million of Tempus/Personalis revenue.

• A $5 billion franchise would imply about 40,000 patients and roughly $360 million of Tempus/Personalis revenue.

• A $10 billion franchise would imply about 80,000 patients and roughly $720 million of Tempus/Personalis revenue.

These are not company projections. They are illustrative. But they make the point.

If this platform spreads across multiple tumor types, the sequencing layer could become a substantial business.

That is what makes the Personalis acquisition more interesting today than it looked a month ago.

Personalis may never be the headline winner. But it could become a sticky part of the system every time one of these therapies is made.

And if Tempus owns Personalis, then Tempus could inherit that strategic position.

The real bull case for Tempus

The bull case is that Tempus becomes a key infrastructure layer in personalized oncology.

This type of treatment is inherently data-intensive. Each patient is a fresh workflow. Each tumor has to be read. Each therapy has to be built from that information.

That creates recurring demand for the data layer.

In a normal companion diagnostic model, a patient may get tested once to determine whether they are eligible for a drug. After that, the test is done.

This is different.

Here, the test is not just a gatekeeper. It is part of the therapy creation process itself.

That makes the workflow potentially more durable and more valuable.

If Merck and Moderna can take this platform from melanoma into lung, bladder, kidney, and other cancers, then the company sitting inside the genomic workflow could end up with a very attractive position.

That is why this story matters for Tempus, even though Tempus is not the name in the breakthrough headline.

Why Iovance belongs in this conversation

There is another company that helps validate the broader concept here: Iovance

Iovance’s AMTAGVI is not a vaccine. It is a tumor-infiltrating lymphocyte, or TIL, therapy. The process is much more labor-intensive.

The patient’s tumor is removed. Immune cells are harvested from the tumor. Those cells are expanded outside the body. The patient then undergoes lymphodepletion, receives the cells back, and is supported with IL-2.

This is a far more operationally heavy treatment model than a personalized mRNA therapy. It involves surgery, cell handling, manufacturing, conditioning therapy, and inpatient-style complexity.

So why does Iovance matter?

Because it is proof that personalized immune-based treatment built around a patient’s own tumor biology can work in melanoma.

That does not mean AMTAGVI and V940 are the same thing. They are not. One is a cellular therapy. The other is a personalized mRNA neoantigen therapy.

But they share a core idea: the patient’s tumor contains information the immune system can be trained or enabled to use.

Iovance helped prove that patient-specific immune therapy is not science fiction. The challenge is that its approach is hard to scale.

That is why Merck and Moderna’s success may be so important.

V940 may offer a path to a more scalable version of personalized immunotherapy.

If Iovance demonstrated that this broader therapeutic concept can produce real results, Merck and Moderna may now be showing how it can become a larger commercial platform.

The bottom line

The market will naturally focus on Moderna and Merck after this Phase 3 result. That makes sense. They are the companies driving the drug.

But smart investors should also pay attention to the supporting cast.

Personalis appears to sit in the tumor-profiling layer behind the therapy. Tempus is in the process of acquiring Personalis. If this melanoma breakthrough turns into a multi-cancer platform, then Tempus could own a valuable part of the picks-and-shovels infrastructure behind it.

That does not mean Tempus deserves to be valued like a drug company. It probably should not be.

But it does mean the Personalis acquisition now looks more strategic than it did before the Phase 3 result.

The deeper lesson is this:

This is not just a melanoma story. It may be the beginning of a new platform for personalized cancer treatment.

And if that platform expands, the winners may include not just the companies making the medicine, but also the companies supplying the genomic intelligence that makes the medicine possible.

That is why this breakthrough deserves a closer look.

It may be the first big commercial signal that personalized cancer immunotherapy is moving from a promising idea into a repeatable system.

And if that is true, Tempus and Personalis may end up mattering more than the market currently thinks.

1) Your life is your reality

2) God has given you certain gifts. You may not know what these are yet, but you have them

3) Your gifts and your actions can, if you let them, shape your reality

4) Therefore, your reality is in your hands

If you don’t believe this, especially if you blame others for your lot in life, you should not be trading. Trading is like golf. It’s you against you. It’s you against ALL OF YOU. If you have an unhealthy view of life, your trading will suffer.

Work on your optimism. I’m not saying naively believe everything in the stock market. But science is awesome! Laws of the universe, we are still discovering, have been harnessed by some of the best engineers, scientists and management teams in the United States of America. There is proof of their work. Look at all the amazing products: the iPhone, EVs, Satellites, Ai, Robotics. Don’t doubt everything as social media has taught you.

And now we have Artificial Intelligence which, to me, is the great accelerator. If you thought the world was moving fast now, look out. Since I started trading in 1996, I have noticed market cycles have been speeding up. I have a feeling Ai is going to accelerate these cycles even more. Embrace it. Ai has helped me immensely in the research piece of trading which is the most time consuming piece.

Anyway, I will get off my soapbox. My point is, you have to be an optimist to do well in trading. We need the Michael Burry’s of the world, but I can tell you from experience, if you go that route in trading, you’re signing up for one of the hardest jobs in the world.

So, in conclusion, a traders mindset should be the following:

1) Optimistic

2) Always Searching for Opportunity… Always

3) Press winners when they are working

4) Cut losers almost immediately. Risk management is #1

5) Have doubts/caution when the market trades under the 50 day moving average

6) Be in cash when the market trades under the 200 day moving average

7) Focus on winners being at least 2.5x losers. That way, even a 40% win rate is a profitable system

Portfolio Update

I do not know whether the current pullback will end next week, develop into a deeper correction or prove to be the beginning of something more serious. No one else knows, either. Markets do not distribute advance copies of their plans.

However, I do believe I understand why the market has been struggling since June.

Identifying the pressures weighing on stocks does not tell us precisely where the indexes will bottom, but it gives us a framework. When one of those pressures begins to ease, we can recognize that an important weight has been removed. When several improve together, the environment can change quickly.

Purely technical traders may not care about any of this. They watch price, volume and trend, and they allow the market to deliver the message. I have always had difficulty completely separating price action from the economic and corporate developments behind it. Understanding the narrative helps me interpret what I am seeing on the screen.

Since early to mid-June, the message has become increasingly clear: the AI bull market needed a break, and perhaps a recalibration.

Here are the major crosscurrents I see.

1. The AI Trade Became Extended

The simplest explanation is often the most important one.

AI-related stocks experienced an extraordinary run. Leadership expanded from GPUs into memory, networking, optical components, servers, cooling, power generation, electrical equipment, rare earths, semiconductor equipment, cybersecurity, neoclouds and space-based infrastructure.

Eventually, too much capital began chasing too many versions of the same theme.

The recent damage has been most visible in semiconductors. After reaching record territory in June, the semiconductor group entered a sharp correction in July. Even after that decline, many of its largest winners remained dramatically higher for the year. That is the profile of an extended leadership group undergoing an unwind—not necessarily a broken secular story.

Long-term bull markets regularly go through periods in which price moves ahead of fundamentals. The correction that follows allows earnings to catch up, speculative positions to clear and investors to distinguish durable growth from enthusiasm.

That appears to be happening now.

The market is no longer rewarding every company with an AI connection. It is beginning to demand evidence of revenue, margins, cash flow and defensible competitive positioning. This is a healthier situation, even when the repricing process is painful.

2. The Memory Trade Became Crowded

Memory was one of the strongest and most crowded areas of the AI trade.

Micron, SanDisk and South Korea’s SK Hynix became major beneficiaries of tight supply, expanding AI infrastructure and rising demand for high-bandwidth memory. Scarcity supported pricing, margins and earnings expectations.

The complication is that higher memory prices eventually become higher costs for customers.

Expensive memory can pressure server economics, increase the cost of AI systems and complicate project-level returns. The same pricing environment that benefits memory manufacturers can reduce profitability elsewhere in the AI supply chain.

South Korea added another layer of instability. Its stock market experienced an extraordinary technology-led advance, followed by a violent reversal in late June. The KOSPI fell nearly 10% in one session after having risen more than 90% for the year, with Samsung Electronics and SK Hynix among the stocks caught in the selloff.

South Korea had become a concentrated expression of the global memory and semiconductor trade. Once leverage and overcrowding began unwinding there, the pressure naturally spread to related stocks in the United States and Japan.

Memory demand may remain structurally strong. The stocks still had to absorb excessive positioning, elevated expectations and the possibility that customers eventually push back against escalating component costs.

3. The Market Is Revisiting the Overbuild Question

The persistent concern surrounding AI infrastructure is whether the industry is building too much capacity.

I do not currently see convincing evidence of a broad demand collapse. Token consumption continues to expand, model usage is rising and AI competition is spreading across the world. The race is no longer confined to a handful of American companies and China.

France has Mistral. The United Kingdom, Canada, South Korea and India are developing their own models, infrastructure and national AI capabilities. Chinese developers continue to introduce increasingly capable open-source systems. More models generally create more inference demand, and inference can become a much larger computing market than training.

Nevertheless, investors remember the telecommunications and internet infrastructure boom of the late 1990s. Enormous capital expenditures were justified by a trend that ultimately proved real, but too much capacity was built too quickly and financed too aggressively.

AI could follow a more productive path and still experience pockets of overcapacity.

The risk is unlikely to appear everywhere simultaneously. It may emerge within particular geographies, chip categories, server configurations or data-center markets. A shortage can become a surplus surprisingly quickly when every company reacts to the same forecast.

The market is therefore looking for evidence that utilization, customer demand and economic returns are keeping pace with construction.

Bookings alone will no longer settle the debate. Investors want to see deployments, usage, revenue and measurable returns.

4. The Iran Conflict Has Reintroduced an Inflation Shock

The conflict involving Iran and the United States has moved beyond a background geopolitical concern. It has become a direct market variable through oil, shipping and inflation expectations.

As of July 19, West Texas Intermediate crude was approaching $85 per barrel and Brent had moved above $90 as fighting intensified and commercial traffic through the Strait of Hormuz faced renewed disruption.

Higher oil prices affect the market through several channels.

They raise transportation and production costs. They pressure consumer purchasing power. They complicate the Federal Reserve’s inflation outlook. They can also keep longer-term interest rates elevated, placing additional pressure on high-multiple growth stocks.

Research assessing the conflict’s inflation effects has estimated that a prolonged disruption could add approximately 0.6 percentage points to U.S. headline inflation during 2026, with a smaller but still meaningful effect on core inflation.

The uncertainty is almost as damaging as the price increase itself. Companies have difficulty planning around a commodity that can move several dollars based on a military headline.

A durable ceasefire, improved shipping conditions or a meaningful decline in oil would remove one of the market’s most immediate macroeconomic pressures.

This is my opinion obviously, but I believe the IRGC will never give up its nuclear ambitions, nor its grip on the country. It will take an invasion by a coalition of the US and Middle Eastern allies (who are tired of having their refineries bombed) to dislodge the IRGC, but if this ever happens, it would definitely come AFTER mid-term elections. Until this happens, it is looking more and more like higher oil prices could become a more permanent headwind.

5. AI Token Economics Are Becoming Deflationary

AI models are becoming more capable, but the price of intelligence is falling.

Chinese open-source developers—including DeepSeek and Moonshot AI—have intensified competition around model efficiency and cost. New models are increasingly capable of delivering strong performance without matching the enormous development and infrastructure budgets of the largest American laboratories.

This creates an important tension.

Lower token prices stimulate usage. Cheaper intelligence should expand the addressable market, accelerate application development and increase the number of workloads that can economically incorporate AI.

At the same time, falling token prices challenge the business models of premium laboratories carrying enormous infrastructure commitments and operating costs. So, I do believe its getting harder and harder for Anthropic and OpenAi to have successful IPOs as LLM competition expands.

The central question is no longer whether AI will be widely adopted. It almost certainly will. The question is where durable economics will accumulate.

Will the largest gains remain with model developers? Will they migrate to cloud platforms, proprietary data, software applications, cybersecurity, robotics or companies using AI to transform existing industries?

Commoditization does not destroy an important technology. It changes where the profits are earned.

That migration of value here is likely to become one of the defining investment questions of the next several years.

6. Seasonality and Positioning Are Less Supportive

Seasonality should never be used as a stand-alone market forecast, but it can influence the character of trading.

July has historically been a constructive month for equities, while August often brings thinner liquidity as institutional activity slows. After an unusually strong advance, reduced liquidity can magnify selling and make individual stocks more vulnerable to air pockets.

The political calendar may also influence risk appetite as the country moves closer to the 2026 midterm elections. Markets frequently react to rising uncertainty surrounding taxes, regulation, spending and congressional control before the results are known.

None of this determines the market’s direction.

It does, however, create a less forgiving environment for crowded positions and companies carrying elevated expectations.

Take a look at this tweet by Wayne Whaley on X. Wayne outlines how the market performs AFTER mid-term elections, and it really is quite good. The period leading up to mid-terms is not very good.

7. The Hardware Trade Is Running Out of Untouched Themes

Investors have already discovered most of the obvious AI infrastructure beneficiaries.

The market has produced extended rallies in:

• GPUs and CPUs

• High-bandwidth memory

• Optical networking and photonics

• Servers and storage

• Data-center cooling

• Power generation and electrical equipment

• Semiconductor manufacturing equipment

• Rare earths and critical minerals

• Neoclouds

• Cybersecurity

• Space and communications infrastructure

There may be additional upside in several of these groups, but they are no longer undiscovered.

Future returns will depend more heavily on earnings execution and less on multiple expansion. Investors will have to identify companies whose profits are growing faster than the expectations already embedded in their stocks.

This is one reason I am becoming more interested in AI’s expansion into the physical world.

Machine vision, industrial automation, autonomous systems, lidar, edge inference and robotics represent a potential next stage of value migration. AI can generate enormous economic value when it moves beyond producing text and images and begins improving factories, logistics networks, vehicles, warehouses and physical processes.

8. Data-Center Permitting Could Become the Ultimate Bottleneck

The AI industry has spent years discussing GPU supply, memory supply, electrical equipment and access to power.

Local political consent may become just as scarce.

Communities across the United States are increasingly aware of the land, water, electricity, noise and environmental demands associated with large data centers. Opposition is becoming organized, national and politically relevant.

On July 18, opponents conducted 142 demonstrations across 42 states. A Reuters/Ipsos poll found that only 14% of respondents supported having a data center located in their community.

Research also suggests that water capacity could become a meaningful infrastructure constraint. Depending on growth and efficiency assumptions, U.S. data centers could require hundreds of millions of gallons of additional daily water capacity by 2030.

The industry can solve many technical bottlenecks by spending more money.

Permitting is different. A project can have financing, customers, chips and a power agreement and still be delayed by local opposition.

The companies most capable of navigating this environment may gain a meaningful advantage. Closed-loop cooling, lower water usage, onsite power, better community agreements and more thoughtful site selection could become increasingly valuable.

What Could Go Right?

The answer is simple: earnings.

Earnings can overwhelm nearly every concern discussed above if companies demonstrate that AI investment is producing revenue, operating leverage and free cash flow.

The market is willing to tolerate enormous capital expenditures when returns are visible. It becomes much less patient when spending rises while monetization remains vague.

This quarter is particularly important because it may determine the next round of AI winners.

The first phase of the trade rewarded infrastructure scarcity. Investors bought the companies supplying the chips, memory, networking, power and cooling required to build AI capacity.

The next phase must provide evidence of economic output.

We need to learn:

• Which hyperscalers are earning acceptable returns on AI capital expenditures?

• Which semiconductor companies retain pricing power?

• Are optical and networking bottlenecks getting better or worse?

• Are power and cooling orders continuing to accelerate?

• Which software companies can convert AI features into higher revenue?

• Are enterprises moving from AI testing into production?

• Which industries are achieving measurable productivity gains?

• Is inference demand absorbing the infrastructure now being built?

If the answers are constructive, the market can begin looking through the current correction.

Dell recently reported quarterly AI-optimized server revenue of $16.1 billion, up 757% from the prior year, and exited the quarter with an AI backlog above $50 billion. Those numbers suggest that infrastructure demand has not simply disappeared.

Vertiv reported 83% growth in adjusted earnings per share during its first quarter and raised its full-year guidance, providing another indication that power and cooling demand remained strong heading into this reporting season.

These are the kinds of data points the market needs.

Strong earnings cannot prevent every correction. They can establish a fundamental floor beneath the companies executing best. Some of my best trading entries have come off sizable earnings gaps. Watch for classic gap and go scenarios.

My Earnings Watchlist

I am particularly interested in year-to-date leaders that produced strong previous earnings reactions, have since pulled back and continue to show relative technical strength.

I am not treating every name below as an immediate purchase. This is a research list. I want to see which companies confirm their prior strength and which ones reveal that expectations have moved too far ahead of results.

AI Infrastructure and Manufacturing

Flex — FLEX

Flex has evolved from a traditional contract manufacturer into a strategic supplier serving data-center, power and other high-growth markets.

Its latest fiscal year ended with fourth-quarter sales rising 17% to $7.5 billion. Adjusted operating margin remained above 6% for the sixth consecutive quarter, demonstrating that the story involves more than revenue growth.

The upcoming report should help determine whether data-center demand remains strong enough to support continued margin improvement.

Dell Technologies — DELL

Dell is one of the cleanest tests of AI server demand.

The company’s last quarter included $24.4 billion of AI orders, $16.1 billion in AI server revenue and an increased full-year AI server revenue forecast of $60 billion.

The major issue is no longer demand alone. Investors will be studying margins, component costs, supply availability and the profitability of the enormous backlog.

Vertiv — VRT

Vertiv remains one of the best-positioned suppliers of cooling and power-management infrastructure for data centers.

Its first-quarter results included rapid earnings growth and a full-year guidance increase. The next report will test whether order growth, pricing and backlog remain strong as investors become more concerned about the pace of data-center construction.

GE Vernova — GEV

GE Vernova sits at the intersection of electricity demand, grid investment and power-generation scarcity.

The thesis extends beyond AI, but data-center load growth has increased the urgency of grid modernization and dependable generation. The important signals will be order growth, backlog quality, equipment pricing and evidence that projects are converting into revenue on schedule.

Networking, Optical Components and Connectivity

Marvell Technology — MRVL

Marvell offers exposure to custom silicon, data-center connectivity and advanced networking.

Its fiscal first-quarter report should be evaluated through the durability of data-center growth, customer concentration and the pace at which custom AI programs move into production. Marvell has also warned that large customers may develop more of their own silicon, making execution and product differentiation especially important.

Lumentum — LITE

Lumentum supplies optical and photonic technologies used in AI and cloud infrastructure. It also entered a multiyear strategic agreement with Nvidia to develop advanced optics for future AI systems.

The market will be watching capacity expansion, customer demand, margins and whether optical requirements continue growing as AI clusters become larger.

Coherent — COHR

Coherent provides exposure to optical communications, lasers and advanced materials.

The central question is whether AI-related optical demand can continue improving the company’s growth and margin profile while offsetting volatility in its other end markets.

MaxLinear — MXL

MaxLinear is a smaller and more volatile way to participate in connectivity and communications silicon.

For MXL, I want confirmation that the improvement is broad enough to become a durable earnings recovery rather than a short-cycle rebound.

Bandwidth — BAND

Bandwidth is different from the hardware names because it provides enterprise cloud communications infrastructure.

Its first quarter produced record revenue of $209 million, up 20%, and the company raised its full-year outlook. Management also highlighted accelerating AI-driven platform adoption.

The next report should reveal whether AI agents and automated communications are beginning to create sustained usage growth.

Semiconductors and Analog Infrastructure

Texas Instruments — TXN

Texas Instruments is less dependent on the AI narrative than many names on this list.

That may be useful. TXN can provide information about the broader semiconductor cycle, including industrial demand, inventory normalization and whether weakness outside AI is beginning to improve.

A stronger analog recovery would broaden the market beyond the most crowded AI trades.

Software and AI Monetization

Datadog — DDOG

Datadog may help answer a central question of this earnings season: can AI infrastructure spending translate into software revenue?

AI applications create more workloads, more complexity and greater demand for monitoring and observability. The bull case requires that usage growth, customer expansion and AI-native workloads produce accelerating revenue without excessive sales or infrastructure costs.

Software may be one of the most important destinations for the next wave of AI value migration.

Building a Robotics Watchlist

I am also beginning to devote more research to companies positioned for AI’s expansion into the physical world.

This is still an early watchlist rather than a finished portfolio.

Cognex — CGNX

Cognex is a leading machine-vision company. Its systems allow factories and logistics operations to inspect products, read codes, guide equipment and automate quality control.

Machine vision is one of the most practical bridges between AI software and physical automation. I want to see whether improving manufacturing conditions and greater AI capability can accelerate growth after a difficult industrial cycle.

Ouster — OUST

Ouster provides digital lidar sensors used in industrial automation, smart infrastructure, robotics and autonomous systems.

The opportunity is substantial, but lidar stocks have historically been difficult to own. Competitive intensity, cash consumption, customer concentration and long commercialization cycles have repeatedly disappointed investors.

For Ouster, revenue growth must eventually translate into credible margins and cash flow.

Ambarella — AMBA

Ambarella develops low-power computer-vision and edge-AI semiconductors.

The stock is exceptionally volatile, but the company has potentially valuable exposure to cameras, autonomous systems, robotics and machine perception. The key issue is whether design wins develop into production revenue at sufficient scale.

Why This Earnings Season May Decide the Next AI Winners

The easiest phase of the AI trade may be behind us.

For several years, investors could buy nearly any credible supplier of scarce AI infrastructure and benefit from rising estimates, expanding valuations and a shortage narrative.

The market is now asking harder questions.

Who is generating an economic return from AI?

Who is capturing revenue rather than merely announcing partnerships?

Who has pricing power after supply improves?

Which companies can grow without consuming unlimited capital?

Which products remain differentiated as models and tokens become cheaper?

These questions will determine where value migrates next.

It may move toward software companies that can charge for AI features. It may shift toward enterprise data, cybersecurity, automation and robotics. It may remain with select infrastructure providers whose bottlenecks prove more durable than expected.

The earnings winners from this quarter deserve unusually close attention.

I will be watching not only whether companies beat estimates, but how stocks respond. A company that reports strong numbers, raises guidance and holds its gains in a difficult market is sending an important signal. A company that produces apparently good results and still sells off may be telling us that expectations remain too high.

Price reaction is part of the fundamental evidence.

Final Thoughts

The current pullback is being driven by a combination of extended positioning, a memory-stock unwind, renewed overbuild concerns, higher oil prices, geopolitical risk, falling AI token prices, seasonal liquidity and growing resistance to data-center development.

Those are legitimate concerns.

They do not automatically invalidate the AI investment cycle.

The more important question is whether the cycle is beginning to produce measurable returns for the companies spending the money. This earnings season should give us the clearest answer yet.

If revenue, margins and cash flow demonstrate that AI investment is paying off, the market can absorb a considerable amount of macroeconomic and geopolitical uncertainty. Investors will return, but they may not return to exactly the same stocks.

Leadership will become more selective. Value will migrate toward the companies converting AI demand into durable earnings.

That is why I ultimately believe this pullback will produce opportunity.

The task now is not to predict the exact market bottom. It is to track the earnings winners, study their price reactions and identify where the next phase of AI value creation is beginning to emerge.

This quarter could provide the map.

Now, check the graphic below when comparing Anthropic Fable costs to GLM-5.2 costs. 1,000,000 output tokens is 11.4 times lower in price. Um sorry, but how does a non-government/non-defense company CFO justify spending 11.4 times more to get “great” Ai when “very good” is sitting right there??

On top of this, GLM-5.2 scores very high in coding which is probably the most desirable skill for an LLM right now. Its just behind Claude Opus and actually ahead of GPT-5.5.

WHAT COULD HAPPEN?

Let’s think about what could happen here. Also, let’s take into account what we are already seeing the U.S. government beginning to do: that is, meddling in LLM software release cycles.

Hypothetical scenario: the “AI sovereignty shock”

Phase 1 — 2026: Model-price arbitrage begins

By late 2026, U.S. companies realize something uncomfortable: for many practical workflows, the best Chinese open-weight models are “good enough” and dramatically cheaper.

The first adopters are not banks or defense contractors. They are:

• coding-agent startups;

• internal enterprise automation teams;

• outsourced software shops;

• data-engineering firms;

• marketing/content automation platforms;

• small SaaS companies trying to preserve gross margin.

They start routing 60–80% of commodity tasks to GLM, DeepSeek, Qwen, or MiniMax-style models. U.S. frontier models remain the “gold tier” for high-value reasoning, but the blended AI bill falls by 50–75%. (Note: below, we are already seeing falling prices in the Token Expenditure Index which measures the blended price of 1,000,000 tokens)

This creates a brutal market message: frontier intelligence is valuable, but not always worth frontier pricing.

OpenAI and Anthropic face a margin squeeze. Their best models cost more to train, more to serve, and now have slower release cadences because the U.S. government is reviewing or influencing access to top-tier models. Meanwhile, Chinese labs release permissive MIT/open-weight models at a faster cadence. The American labs are stuck between three bad choices:

1. Cut API prices hard, compressing margins.

2. Keep prices high, losing enterprise volume.

3. Restrict access for safety, allowing open Chinese alternatives to gain share.

Phase 2 — 2027: CIOs normalize “model routing”

The investment world initially treats this like a software margin story. Analysts start asking: “Are OpenAI and Anthropic actually platform monopolies, or are they high-cost research labs whose product gets commoditized every six months?”

Large enterprises do not necessarily “switch to China.” They build routing layers:

• U.S. model for sensitive data.

• Chinese open model for coding drafts.

• Local self-hosted model for document extraction.

• Cheapest acceptable model for customer service.

• Frontier model only when accuracy/reasoning stakes justify it.

This is exactly where the pricing pressure becomes dangerous. If Anthropic or OpenAI loses the high-volume middle layer, they still have prestige, but they lose the usage base that funds the next generation.

Public-market investors then start repricing the sector. Instead of valuing frontier labs like software monopolies, they value them more like capital-intensive defense contractors plus cloud-scale R&D labs. Multiples compress.

OpenAI and Anthropic delay or downsize IPO plans. They can still raise money, but each raise becomes more strategic: Microsoft, Amazon, Google, sovereign wealth funds, defense primes, and the U.S. government all become possible financing partners. Reuters has already reported that both OpenAI and Anthropic have confidentially filed for U.S. IPOs, and separately that policymakers have discussed public stakes, taxes, board representation, or funding-for-equity structures for AI firms.

Phase 3 — 2027: Washington sees a national-security paradox

This is the moment the U.S. government panics.

The paradox is:

The U.S. is slowing or gating its own frontier models for safety, while China releases powerful open models cheaply and globally.

From Washington’s perspective, that creates three risks:

1. Capability leakage: U.S. firms become dependent on Chinese-origin models embedded deep inside software workflows.

2. Security uncertainty: even if weights are open, enterprises may use Chinese APIs, hosted endpoints, or opaque fine-tunes.

3. Economic hollowing: the most strategically important U.S. AI labs cannot monetize enough volume to fund the next model generation.

The government’s first response is not nationalization. It is procurement and compliance.

A new rule appears: any AI used in critical infrastructure, defense supply chains, financial-market infrastructure, healthcare claims systems, or federal contractors must use a Trusted Frontier AI Provider or a certified domestic/open model with audited provenance.

This is sold as “cybersecurity,” but the effect is industrial policy.

Chinese models are not banned everywhere. They are gradually pushed out of sensitive enterprise workflows. Startups can still use them. Consumer apps can still use them. But regulated enterprise demand gets steered back toward U.S.-approved labs.

Phase 4 — 2028: The “American Frontier AI Compact”

By 2028, OpenAI and Anthropic are too important to fail but too expensive to support purely through market pricing.

The government creates something like an American Frontier AI Compact. It has four pillars:

1. Guaranteed federal demand.

The U.S. commits multiyear procurement to approved U.S. frontier labs for defense, cyber, intelligence, IRS modernization, VA healthcare, procurement fraud, software modernization, and national lab research.

2. Compute financing.

Instead of giving OpenAI/Anthropic blank checks, the government guarantees loans for U.S.-based AI data centers, similar in spirit to strategic infrastructure financing.

3. Equity or warrants.

The Intel precedent becomes the template. The U.S. government already converted CHIPS-related funding into a passive equity stake in Intel, with Intel saying the government’s stake was funded by remaining grants and Secure Enclave awards.

4. LLM Release governance.
Frontier model releases above a capability threshold must go through a 30-day federal review, red-team testing, and staged release to trusted partners. This formalizes what is already beginning to happen with GPT-5.6 and Anthropic’s Mythos/Fable access controls.

At this point, OpenAI and Anthropic are still private companies, but their strategic freedom is reduced. They look less like pure software companies and more like Bell Labs or “cloud utilities”.

Phase 5 — 2029: The crisis that triggers quasi-nationalization

The trigger would probably not be “Chinese models are cheaper.” That alone is not enough.

The trigger would be a combined shock:

• A major U.S. bank, exchange, hospital network, or energy operator is found to have deployed Chinese open models in mission-critical automation.

• A cyber incident is blamed, fairly or unfairly, on unsafe model provenance.

• Anthropic or OpenAI faces a liquidity crunch because API margins collapse and release restrictions reduce revenue.

• A foreign investor or hyperscaler tries to acquire too much control.

• Congress decides frontier AI is a national strategic asset like semiconductors, nuclear technology, or telecom infrastructure.

Then the government steps in with a rescue/compact.

Not “we own 100% of OpenAI.” More like:

• U.S. government receives 10–25% preferred equity or warrants.

• A federal board observer is installed.

• Certain model weights, safety systems, and data-center capacity are deemed strategic assets.

• Foreign ownership is capped.

• Government gets emergency access rights during cyber/national-security crises.

• Approved labs receive guaranteed compute financing and federal purchasing commitments.

• Labs agree to staged release rules and domestic-security requirements.

This is not classic nationalization. It is regulated strategic dependency.

Would this become a U.S. Department of Artificial Intelligence?

Eventually, maybe—but the first version would probably not be cabinet-level, but we could eventually, get there. A slow nationalization or slow merge into a “cloud utility” seems to me a likely outcome.

Reviewing these charts, I noticed many of these leaders had quite “smooth and strong” trends in April. After all, the market pivoted up in April and after a multi-month down period, the first up month is usually the strongest. As an aside, the more I trade, the more I realize how important a pivot month is. Your entries in that month can set you up for the duration of the new intermediate term bull move. This is an important concept. And in April, in opposition to that concept, I took large positions and scaled out into strength in late April. Well, that was clearly the wrong move as this bull has really become incredibly strong since the April pivot.

So, the strength in May and June has caused me to seek re-entries and these have largely not gone well for me. After reviewing charts, I believe I understand why. Many of the leaders have become VERY volatile in May and June after showing smooth and strong trends in April. Here’s some examples:

AEHR is the poster child for “strong in April” and complete chaos thereafter. The high volume in April gave us the strong and smooth uptrend. Essentially there were very few sellers and a TON of eager buyers. That condition always results in a strong bull move.

AAOI was quite strong in the first half of April, but since then, it’s become a Rorschach inkblot test, literally impossible to trade UNLESS you had a much earlier entry (April or before) which affords you the luxury to hold through this mess.

There have also been a LOT of episodic pivots like the MXL chart. These are tricky to trade, but if you recognize it as an EP early, and can buy near the bottom of the gap, these trades can turn out well. But even MXL gave you 9 really strong trading days after the EP, and then got really chaotic.

So where am I going with all of this? In a trading regime that has gone from smooth and strong in April to quite chaotic, your entries need to be pinpoint accurate or you will get stopped out. This is what has happened to me as of late; my entries have not been good enough.

Good Scanning Can Help

Good scanning can help with this problem. FinViz is my scanning tool of choice. TC2000 and TradingView are also solid. For the purposes of this article I will show you the scans I’ve created in FinViz. Before we start reviewing those, I wanted to review this BE chart because it’s a great example of probably one of the best setups in this current environment, the Failed Breakdown:

I have seen this failed breakdown across multiple stocks. What does it mean? Essentially there is big money waiting to buy when the up trend looks MOST in jeopardy. So, as soon as the stock dips below 20 or 50 day lows (usually seen with trend line resistance) buyers step in and send the stock on another leg of the bull run.

So, with the BE setup in mind, I set up the following FinViz scan:

Failed Breakdown Scan

Here’s the specifics:

1. Market cap: stocks > $2 billion in market cap

2. Price > the 200 Day Moving Average

3. Change from the Open = Up

4. 0% - 10% Below the 20 Day Low

5. Current Volume > 1,000,000 shares

This will give you liquid stocks trading above their respective 200 day moving average (healthy) that are 0% to 10% below the 20 day low. And, change from open = up gives you some buying in the current day. This scan would have given you BE on the day of the Failed Breakdown.

Scan Against Distance from 20 or 50 Day Moving Averages

Another solid scan that can help you identify year to date leaders that have pulled back, is as follows:

Here are the specifics:

1. Market cap > $2 Billion

2. Average Volume > 1,000,000

3. Performance Year To Date > 50%

4. -10% to +5% Distance from 20 Day Moving Average

The key for this scan is that it gives you ytd leader candidates that have pulled back to the 20 day moving average, either a bit below the moving average or slightly above. Therefore, in a chaotic stock, you are not buying high where you can easily get stopped out. If you want a deeper pullback, you can expand the percentages around the 20 day, or, utilize similar parameters around the 50 day moving average.

To sum up, this market has not been easy because many of the leaders have been exceptionally volatile in May and June. You can help your trading by consistently scanning for leaders that have pulled back to key areas where you have less of a chance of being stopped out.

And by the way, for you OpenClaw users, FinViz has an API (which I’ve been utilizing) so you can automate your scans and send the output wherever you want. I have created a system that scores stocks using fundamental data from Financial Modeling Prep and then combines results from my FinViz technical scans. This, essentially, gives me stocks with strong fundamentals which are technical buys.

After a long pullback, OKLO is interesting again because the story has changed. A year ago, the market was mostly buying a vision: small advanced nuclear reactors, AI data centers, Sam Altman, and the need for clean baseload power. Today, the company is starting to show actual milestones across three areas: power, fuel, and isotopes.

That is the key. Oklo is not just trying to build a reactor. It is trying to build a nuclear platform.

The Near-Term Catalyst: July 4 Criticality

The biggest near-term catalyst is Groves, the isotope test reactor being built through Oklo’s Atomic Alchemy subsidiary.

Management has said Groves is targeting criticality by July 4, 2026. In plain English, “criticality” means the reactor achieves a self-sustaining nuclear chain reaction. It does not mean commercial electricity production. It does not mean Aurora is already powering data centers. But it would be a major psychological and technical milestone.

Why?

Because nuclear is viewed as slow, expensive, and endlessly delayed. See the 24% short interest on the stock as proof of this. If Oklo can show that a new nuclear asset can move from construction into startup on an accelerated timeline, investors will immediately extrapolate that progress to the broader platform.

Groves is not the main Aurora power reactor. It is an isotope reactor. But it gives Oklo a real-world proving ground: procurement, construction, regulatory sequencing, fuel delivery, installation, testing, startup procedures, and operations. Those lessons can feed directly into future deployments.

The stock will likely treat a successful Groves criticality event as proof that Oklo is no longer just a slide-deck nuclear company.

The Bigger Prize: Aurora

The real investment case is Aurora, Oklo’s fast-fission powerhouse.

Aurora is designed to deliver small-scale, reliable nuclear power close to where power is needed. That is the exact problem AI data centers are creating. The grid is strained. Interconnection queues are long. Gas turbines are backordered. Renewables are intermittent. Data centers need firm power, 24/7.

Oklo’s pitch is simple: place small nuclear powerhouses near large power users and sell energy under long-term agreements.

That is why the Meta and Switch relationships received so much attention. Investors are not buying Oklo because they think nuclear is interesting. They are buying it because AI infrastructure may require a massive new layer of power supply, and Oklo has positioned itself as one of the few public pure plays on that theme.

Recent Management Commentary

On the most recent management call, Oklo sounded like a company shifting from planning to physical execution.

Management said the company is building across three business lines: power, fuel, and isotopes.

On power, Aurora-INL has submitted its Preliminary Documented Safety Analysis for DOE review. Oklo also received NRC approval for its Principal Design Criteria topical report for Aurora. That is a regulatory building block. It establishes key safety and design criteria that can be referenced in later applications, reducing repeated review.

On fuel, management emphasized that fuel availability is one of the major gating items for advanced nuclear. This is critical. A reactor design is not enough. You need fuel. Oklo is working on multiple paths: fuel fabrication, recycling, potential plutonium-based bridge fuel, and broader fuel partnerships.

On isotopes, management said Groves has its PDSA in review, its DSA submitted, and has received a certificate of substantial completion for construction. The company said the next steps are readiness review, startup approval, final installation of reactor equipment, integrated system testing, and fuel delivery.

That is a very specific checklist. Today, I am seeing a very significant number of July $100 Calls being traded today, and I believe the market is trying to price the possibility that Oklo is close to showing a real nuclear asset coming alive in July.

Why the Pullback Creates Opportunity

OKLO is a high-risk stock, but pullbacks in story stocks can become attractive when three things happen at once:

First, expectations reset. The stock has already gone through a painful decline from its prior hype levels. That removes some of the froth.

Second, the company keeps hitting milestones during the decline. This is the part investors often miss. A falling stock price does not always mean the company is failing. Sometimes the stock is digesting valuation while the business is advancing.

Third, the next catalyst is close. July 4 is not years away. If Groves reaches criticality, OKLO will have a tangible milestone at the exact moment investors are looking for proof that advanced nuclear can actually move faster.

This creates a favorable setup: lower stock price, better balance sheet, clearer milestones, and a near-term technical catalyst.

The Technicals

Right now, the stock is not in the wheelhouse of momentum investors. It has fallen below the 200 day moving average and there is plenty of overhead resistance from the long decline. However, price is staying above the 50 day moving average and higher lows have begun to form. On top of that, price has compressed which is one of my key indicators. Check out the six day tight flag forming now. If this stock can get over near term resistance at $80 and then the 200 day moving average at $85.70, I think it could really get going, especially with a short interest of 24%.

The Balance Sheet Is a Weapon

Oklo ended Q1 with a very large cash and marketable securities position. For a development-stage nuclear company, this is essential.

Most nuclear startups fail long before the technology has a chance to prove itself because they run out of money. Oklo has given itself runway. The company is spending aggressively, but it has capital to pursue multiple workstreams at the same time: Aurora-INL, Aurora-Ohio, Aurora-Eielson, fuel fabrication, fuel recycling, Groves, and isotope commercialization.

That flexibility separates Oklo from smaller nuclear concepts that may need constant financing just to survive.

The Risk

The risk is obvious: Oklo still has to execute.

Groves criticality could slip. Aurora could take longer than expected. NRC and DOE processes can still create delays. Fuel supply remains a central issue. The company is pre-revenue in its main power business. If the stock runs hard into a July event and the event disappoints, the downside could be sharp.

This is not a sleepy energy investment. It is a venture-style public equity.

The Bottom Line

Oklo is attractive now because the company is approaching a moment where investors may get the first real evidence that its nuclear deployment model works.

The July 4 Groves target is the near-term spark. Aurora is the long-term prize. Fuel recycling and fabrication are the strategic moat. Isotopes could provide earlier commercial activity. AI data-center power demand is the macro tailwind.

The market has been waiting for proof that Oklo can move from nuclear story to nuclear execution. The next few months may provide exactly that.

For investors who can tolerate volatility, OKLO is one of the cleanest public ways to bet on the next phase of the AI power buildout: not chips, not data centers, but the energy layer underneath them.

Holdings Update

It’s been really hard to feel comfortable buying into this over extended market, but on rare occasions like this insane bull run, you have to force yourself.

Y’day I scaled into TEM equity, AVAV equity, PLTR Calls and I’m also holding NOK calls. Clearly, I’m extremely bullish on AMZN. I’ve been in and out of NVDA (NVDL) trying to catch the upturn and currently have a small NVDL position. NVDA is STILL not getting rewarded, and as I wrote in my Nvidia Market Cap Bias article, I think big money managers will ultimately be proven very wrong here. This Ai trend can go on until the mid 2030s. In the Dell call last night, the following quote was uttered and I think it’s gold:

“The majority of the installed base remains on 14th-generation or older servers, reflecting the continued refresh opportunity going forward.”

Actually, it wasn’t just the post. It was the conjunction of a few things that started to form a much bigger concept in my head.

First, the company I work for is implementing AI. But we are not necessarily using the newest, most powerful, most expensive frontier models. We are using a “good enough” model.

This means the model does not need to be the smartest model on planet Earth. It does not need to crush every benchmark. It does not need to write poetry, solve physics problems, code a trading platform, and pass the bar exam all at the same time.

It just needs to do the job.

On top of this, “work routing protocols” are starting to pop up. The idea is simple, but potentially very important: software can evaluate the work you are trying to accomplish and route that work to the appropriate LLM.

Not the best LLM. The appropriate LLM.

That means if the task is basic, it goes to a cheaper model. If the task is moderately complex, it goes to a mid-tier model. If the task is truly difficult, then and only then does it get routed to the most expensive frontier model.

And then I saw Deirdre Bosa’s post, which basically pointed to the same trend: companies are realizing they are spending a lot on AI, and they may increasingly shift toward cheaper models that are “good enough” to get the job done.

That stopped me in my tracks.

Because as bullish as I have been throughout this AI bull market, this concept introduces a very uncomfortable question:

What if revenue per token is going down?

And what if that is happening faster than investors realize?

The AI bull case has been built on two assumptions

The AI bull case has a few big pillars.

First, usage will explode.

Second, companies will pay a premium for intelligence.

Third, the best models will command the best economics.

Fourth, the companies sitting closest to frontier AI will capture enormous value.

I still believe the first point is true. AI usage is going to explode. In fact, cheaper AI probably makes usage explode even faster.

But the second, third, and fourth points are where I am starting to pause.

Because if companies start routing work to cheaper models, then the economics change.

The world may use more AI, but pay less per unit of AI.

That is a very different stock market setup than the one many investors are underwriting today.

Revenue per token could fall

This is the key point.

If work routing becomes normal, then the average price paid per token should fall.

Think about it.

Today, a company might send too much work to an expensive model because that is the easiest setup. You pick a model, plug it into your workflow, and use it broadly.

But that is not how cost-conscious enterprises behave over time.

Enterprises optimize.

They ask:

“Why are we using the most expensive model for a simple customer service question?”

“Why are we using frontier AI to summarize a short internal document?”

“Why are we using the smartest model available to classify emails, clean data, draft boilerplate, or answer routine employee questions?”

The answer is: they probably won’t.

At least not forever.

The next stage of AI adoption may be less about who has the most powerful model and more about who can route work to the cheapest model that gets the job done.

That is a huge shift.

And if that shift happens, then yes, revenue per token should decline.

This is bullish for AI adoption, but not necessarily bullish for AI stocks

This is the part that matters most for investors.

A decline in revenue per token does not mean AI is failing.

Actually, it may mean AI is succeeding.

Lower costs mean more use cases become economical. Small businesses can use AI. Consumer apps can use AI. Internal corporate tools can use AI. Edge devices can use AI. Call centers, legal teams, finance teams, healthcare administrators, coders, marketers, and analysts can all use more AI.

So AI adoption could keep ripping higher.

But stock prices do not only care about adoption.

Stock prices care about profits.

And if AI becomes cheaper faster than companies can monetize it, the market may have to rethink who actually captures the value.

That is why this concept bothers me.

Because “AI is everywhere” and “AI companies make huge profits” are not the same statement.

The internet became everywhere.

Consumers won.

Many businesses won.

But a lot of early internet companies did not capture the economics investors expected.

That is the risk here.

OpenAI and Anthropic may not want this discussion before IPO

This is where the timing gets interesting.

OpenAI and Anthropic are reportedly heading toward enormous future valuations. The valuation premise is that frontier AI remains scarce, powerful, and monetizable.

But what if the market is already moving in the other direction?

What if enterprise customers are increasingly saying:

“We like frontier models, but we only need them for a fraction of our work.”

What if the default architecture becomes:

• route simple tasks to cheaper models,

• route medium tasks to mid-tier models,

• reserve frontier models only for the hardest work?

That does not destroy OpenAI or Anthropic.

But it does change the valuation framework.

Instead of being the default intelligence layer for the enterprise, frontier labs could become premium specialists used only when necessary.

That is still valuable.

But is it worth the kind of valuation people are talking about?

That is the question.

And I do not think OpenAI or Anthropic want investors obsessing over “revenue per token” ahead of their IPOs.

Because if investors start asking whether revenue per token is structurally falling, the IPO conversation changes.

The question becomes less:

“How big can AI get?”

And more:

“How much pricing power can the model companies actually keep?”

The biggest problem: model intelligence may become abundant

The market has been treating intelligence like it is scarce.

But what if intelligence becomes abundant?

What if the cutting edge remains expensive, but “good enough” intelligence becomes cheap and widely available?

That would create a deflationary force inside the AI market.

And deflation is usually great for users, but dangerous for companies valued on premium pricing.

If several models can complete the same enterprise workflow, customers will not pay premium prices forever. They will run bake-offs. They will compare outputs. They will compare latency. They will compare accuracy. They will compare security. And then they will compare cost.

Eventually, procurement departments get involved.

And when procurement gets involved, pricing power usually goes down.

That is exactly what work routing accelerates.

It turns AI into a competitive market at the task level.

Not “which model is best overall?”

But:

“Which model is cheapest for this specific job while still meeting our quality threshold?”

That is a brutal question for premium model pricing.

This could hit Nvidia too

The first-order concern is OpenAI and Anthropic.

But the second-order concern is Nvidia and the broader AI infrastructure trade.

To be clear, I am not saying this kills Nvidia. That would be foolish. Nvidia is one of the most important companies in the world, and AI demand is real.

But if the market starts believing that a lot of AI workloads can be handled by smaller, cheaper, optimized models, then investors may start questioning the amount of ultra-premium compute required over time.

The bear case is not:

“AI is fake.”

The bear case is:

“AI is real, but maybe not every task requires frontier-scale compute.”

That distinction matters.

If AI demand shifts from frontier training and premium inference toward cost-optimized inference, then the market may begin to reprice parts of the AI infrastructure stack.

That could pressure:

• Nvidia,

• AMD,

• Broadcom,

• Marvell,

• Arm,

• AI networking names,

• power/cooling infrastructure names,

• and any company being valued as if AI capex grows in a straight line forever.

Again, usage can rise while revenue per token falls.

That is the uncomfortable part.

The software layer may face pressure too

There is another risk.

If model costs fall, customers may also question premium AI software pricing.

Many software companies are trying to charge more for AI features. Some of those features are valuable. Some are deeply embedded into workflows. Some deserve premium pricing.

But others may be glorified wrappers around models that are becoming cheaper by the month.

At some point customers may ask:

“Why am I paying this much for an AI feature if the underlying model cost keeps falling?”

That does not mean all AI software is in trouble.

The winners will be companies with proprietary data, deep workflow integration, distribution, and measurable ROI.

But generic AI wrappers could be vulnerable.

If the model is commoditized, the moat has to come from somewhere else.

The winners may be different than people think

If this trend plays out, the winners may not be the companies with the absolute smartest model.

The winners may be the companies that make AI cheaper, easier, and more efficient.

That could include:

* model routers,

* inference optimization platforms,

* companies with custom silicon,

* open-source ecosystems,

* cloud providers with low-cost AI infrastructure,

* application companies with proprietary data,

* companies that own workflow distribution,

* and edge AI players that move inference closer to the device.

In other words, the market may shift from rewarding “frontier intelligence” to rewarding “cost-effective intelligence.”

That is a major regime change.

My concern

I have been very bullish on AI.

I still believe AI is one of the most important technology shifts of my lifetime.

But this concept is making me think harder.

Because if revenue per token is falling, investors need to be careful about what they own.

AI adoption can boom while certain AI stocks fall.

That is not a contradiction.

It is exactly what happens when a technology becomes more useful but less scarce.

The customer wins.

The economy wins.

Productivity may eventually win.

But the companies priced for permanent scarcity may not win.

The simple thesis

Here is the whole thing in one sentence:

If companies increasingly route work to cheaper “good enough” models, then revenue per token should fall, and the AI market may have to reprice the companies valued as if frontier intelligence will hold premium pricing power forever.

That does not mean the AI bull market is over.

But it does mean the bull market may change.

The next phase may not be about who has the smartest model.

It may be about who can deliver acceptable intelligence at the lowest cost per task.

And if that is where we are going, investors should be prepared.

The stock has been through the software-stock meat grinder. It traded around $279 and then got crushed all the way down to roughly $119. That kind of drawdown usually tells you the market has made a decision. The market decided Snowflake was just another expensive software stock — high multiple, slowing growth, too much hype, not enough operating leverage.

But I’m starting to wonder if that framing is wrong.

Snowflake may be getting lumped into the old software bucket at exactly the moment it should be getting looked at as part of the new breed of AI infrastructure winners.

This is not Nvidia. It is not a chip company. It does not sell GPUs. But it sits in one of the most important layers of the AI stack: enterprise data.

And AI does not work without clean, organized, accessible, governed data. That is where Snowflake lives.

The market may be misclassifying Snowflake

The mistake, in my opinion, is viewing Snowflake as a legacy software name instead of as an AI-enablement platform.

Companies do not just wake up one day and “do AI.” They have to organize their data. They have to govern it. They have to make it usable. They have to connect it to models, applications, agents, dashboards, and workflows.

That is exactly what Snowflake is trying to become.

On the last earnings report, Snowflake showed that its business is still growing at a very healthy pace. Fourth-quarter product revenue was $1.23 billion, up 30% year over year. Remaining performance obligations were $9.77 billion, up 42% year over year. Net revenue retention was still 125%, and Snowflake had 733 customers doing more than $1 million in trailing-12-month product revenue.

Those are not dead-software numbers. Those are strong platform numbers.

The company also guided fiscal Q1 product revenue to $1.262 billion to $1.267 billion, representing about 27% growth, and guided full-year fiscal 2027 product revenue to $5.66 billion, also about 27% growth.

So the question heading into earnings is pretty straightforward:

Is Snowflake still just decelerating software, or is it becoming one of the core AI data platforms enterprises need to build on?

I think the answer may be the latter.

The sell-side is starting to lean bullish

Two analyst notes caught my attention.

First, Mizuho reportedly said its channel checks ahead of earnings were favorable for Snowflake. The firm pointed to healthy data-modernization demand, continued data-warehousing demand, and rising adoption of products like Snowflake Intelligence and Cortex Code. Mizuho has an Outperform rating and a $220 price target on the stock.

Then Bank of America came out with a similarly positive view. BofA raised its Snowflake price target to $205 from $195 and said it expects a strong fiscal Q1 because demand entering the year remains robust. More importantly, BofA said it does not expect the quarter to change its view that Snowflake is gaining share in the AI business-intelligence opportunity.

One bullish analyst note is interesting. Two bullish analyst notes, both pointing to demand strength and AI-related opportunity, starts to look like a real setup.

Now, I do not buy stocks just because analysts like them. We all know how that goes. But when analyst checks line up with company commentary, product momentum, and external interest signals, I start paying attention.

Cortex Code may be more important than the market realizes

The product that really caught my eye is Cortex Code.

Snowflake describes Cortex Code as an AI-driven coding agent integrated into the Snowflake environment. It is designed to help with data engineering, analytics, machine learning, and agent-building workflows. In other words, it is not just a generic AI coding assistant. It is built inside the Snowflake ecosystem.

Snowflake is not only trying to host data. It is trying to become the place where companies build AI-enabled workflows on top of that data.

And here is the key part financially: Cortex Code is billed based on token consumption for customers with existing Snowflake accounts, and we all know how token use has been exploding recently across LLMs. Snowflake also says Cortex Code usage can involve normal Snowflake compute and storage charges depending on the workflow.

That means this product has a direct monetization path.

If usage grows, Snowflake can potentially benefit in multiple ways:

That is the kind of product the market should care about.

Google Trends is flashing an interesting signal

I ran “Cortex Code” through Google Trends, and the result is worth paying attention to. Now, Cortex Code was just released in February of 2026, so this does not prove a whole lot.

Let me be clear: Google Trends does not equal revenue. It does not prove adoption. It does not prove paying customers are using the product at scale.

But it is a signal.

And for a product like Cortex Code, which is tied directly to Snowflake’s AI narrative and potentially to token-based revenue, I think that signal matters.

At minimum, it tells me there is real awareness and curiosity around the product. In enterprise software, awareness usually comes before trials. Trials come before adoption. Adoption comes before revenue.

So no, I would not bet the farm on a Google Trends chart. But I would absolutely include it in the mosaic.

And similarly on Google Trends, the keywords “Snowflake” (specified to refer to the software company) and “Snowflake Intelligence” are both making spike highs over a 5 year period in the first couple months of 2026. These are additional signs of increasing interest.

The earnings setup: revenue beat likely, EPS is the question

Here is where I am on the actual quarter.

I am getting increasingly confident that Snowflake can beat revenue expectations.

Between the company’s last-quarter momentum, strong RPO growth, still-healthy net revenue retention, favorable analyst checks, and the visible interest around Cortex Code, the revenue setup looks good.

My bigger concern is EPS.

Snowflake is still investing aggressively. It is building AI products. It is funding product development. It is expanding its platform. That is good strategically, but it can limit near-term earnings upside.

The company guided Q1 non-GAAP operating margin to 9%, while full-year fiscal 2027 non-GAAP operating margin guidance was 12.5%. That gives Snowflake room to beat, but it also tells me the market will be watching expense discipline closely.

This is the part that gives me pause.

A revenue beat is good.

A revenue beat plus EPS beat is what makes the stock work.

A revenue beat plus margin upside plus confident full-year commentary could make the stock move hard.

Valuation is not crazy versus recent software winners

This is another part of the setup I find interesting.

Two recent earnings winners — Datadog and JFrog — have been rewarded by the market. Their price-to-sales multiples are much higher than Snowflake’s.

By the numbers I’m looking at:

Snowflake is not cheap in an absolute sense. I am not pretending it is a value stock.

But relative to other high-quality software names that are showing growth, Snowflake does not look outrageous. In fact, if Snowflake shows revenue acceleration and proves that AI products are contributing to consumption, I could see the market deciding that 12.74x sales is not expensive enough.

That is the opportunity.

The market has already punished the stock like a decelerating software name. If Snowflake can show it is actually an AI data-platform winner, the multiple can expand.

Innodata Results

Another interesting read-through is Innodata. INOD recently crushed earnings and had a huge stock reaction because the market is starting to reward companies tied directly to the AI data layer. INOD is not Snowflake. INOD helps AI builders with training data, post-training data, model evaluation, alignment, trust and safety, and human-expert workflows. Snowflake is a software platform where enterprise data lives, gets governed, and gets activated into analytics and AI applications.

But the common denominator is important: AI needs data infrastructure.

That is the part of the Snowflake story I think the market may be missing. Snowflake is not just a legacy software stock. It is one of the companies helping enterprises make their data usable for AI.

What I want to hear on the SNOW call

For me, the checklist is pretty simple.

I want to hear management talk about:

1. Product revenue acceleration

2. Cortex AI and Cortex Code usage

3. Snowflake Intelligence adoption

4. AI workloads driving consumption

5. Stable or improving net revenue retention

6. Operating-margin discipline

7. Confidence in full-year guidance, or ideally a raise

The dream quarter would be: product revenue beat, EPS beat, operating margin above guide, and management clearly saying that AI-related workloads are helping drive consumption.

That would change the narrative.

That would support the idea that Snowflake is not old software. It is AI infrastructure software.

Final thought

I am thinking about taking a position ahead of earnings, but I am not exactly sure how I will do this (options or stock). I will let subscribers know what I do in a note on the day of earnings. Earnings are on May 27th after the close of the market.

I’m pretty confident in the possibility of a revenue beat. The company’s last report was strong. Mizuho’s checks sound positive. Bank of America’s note sounds positive. And the Google Trends data is a legitimate signal that interest in Snowflake and in an important Snowflake AI product has picked up significantly.

But the EPS beat is what gives me pause.

Snowflake needs to show that it can grow and still deliver leverage. The market is not going to reward “growth at any cost” the way it did a few years ago. It wants revenue growth, but it also wants margin discipline.

Still, I think the setup is interesting.

The stock has already been crushed from its highs. The market has thrown it into the software bucket. But Snowflake may belong in a different bucket entirely: the companies helping enterprises actually implement AI, like Jfrog, Innodata and DataDog.

And if Snowflake proves that next week, there could be a very nice post-earnings move.