Chartrand’s best sentence is the one most readers will skim. It would be harder, he wrote, to find a billionaire who got rich “through voluntary trade” without making something cheaper or better. Those three words carry the whole sentence. Take them out and the dare has a second answer, older than anything on either list.

The serious version of the dare was made four years ago by David Nasaw, the historian who wrote the standard biography of Andrew Carnegie. His New York Times essay, “The Robber Barons Had Nothing on Elon Musk,” argued that American moguls have always played a winning game with government: railroads built on federal land and paid by the mile, steelmen sheltered by tariffs, and Musk carried by Tesla subsidies and NASA contracts. This week the argument gets a sequel on screen, when Alex Gibney’s documentary “Musk” comes to home video on Friday, three days after a midterm election in which its subject had spent $446 million by late October.

Nasaw is right about the railroads, and the railroads are what the label was built on. The term began as an 1870s farmers’ insult for railroads that lived on land grants. Ida Tarbell gave it a villain in 1904, writing with real documentation and a real grievance: Standard Oil’s schemes had ruined her father and brother. By 1934, when Matthew Josephson’s book made it canonical, the word covered everyone who had built anything large.

The test that sorts them is one a reader can check. Congress chartered the Union Pacific and the Northern Pacific, gave them land and lent them money by the mile. James J. Hill built the Great Northern to Seattle without a federal land grant, on a Minnesota line that carried a state grant and on private capital. When the Panic of 1893 hit, the subsidized transcontinentals went into receivership; the Union Pacific’s own history records that it “was placed in the hands of three receivers.” The Great Northern did not. Historian Burton Folsom gave the difference names: market entrepreneurs, who got rich selling to customers, and political entrepreneurs, who got rich selling to legislators.

Cornelius Vanderbilt ran the same experiment at sea. In 1847 Congress paid Edward Collins $385,000 a year to carry mail to Liverpool. When the line lost money, Collins lobbied for more and got $858,000, while Vanderbilt, with no subsidy at all, cut costs and fares. Congress withdrew the subsidy, and the Collins Line stopped sailing in 1858.

SpaceX is the Vanderbilt story under a different flag. In 2014 NASA hired two companies to carry astronauts to the space station, on fixed-price contracts and the same set of requirements. Boeing asked $4.2 billion and SpaceX asked $2.6 billion. In March 2025, two astronauts who had flown up on Boeing’s Starliner came home on a SpaceX Dragon. Nasaw counts that contract as a subsidy. The contract is closer to a sale: NASA was the customer, paid less and got the ride. (I made a version of this argument in “The Tollbooth and the Giveaway.”)

The same test does not let Musk off. From 2009 through 2023, Tesla collected almost $9 billion selling regulatory credits, permits that exist because California, the European Union and China require carmakers to sell electric cars and let them buy compliance from someone who does. None of it is illegal. Collins broke no law either; Congress wrote him the check. The test is not whether the money is legal but who chose to pay it. That money ends up in somebody’s sticker price. Spending $446 million to elect legislators is not a subsidy either, but it is the move Folsom’s political entrepreneurs made first. The honest verdict on Musk is a split one: the rockets were sold to a customer, the credits to a mandate, and the checkbook sits on the Collins side of the ledger.

So why does the label stick to the builders? Economist Deirdre McCloskey’s answer is that the way a society talks about merchants decides how much room it gives them, and that talk has a built-in bias. A ruined competitor has a name, a town and a daughter who writes. The millions who bought cheaper kerosene, cheaper freight and cheaper rides to orbit have no story, because a price that falls is not an event. The distinction is older than the railroads. For eight centuries, the people who built commercial order were the ones who served customers, and the people who got rich from a charter were the ones who earned the label.

Back to the dare. Name one time a billionaire lowered your bills. The ones who sold to you did it every time you filled a lamp or bought a ticket. The ones who sold to legislators sent you part of theirs, and the invoice never had their name on it.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

He has tried hard to find out. In 2023, Party inspectors punished 49 officials of provincial or ministerial rank and 24,000 at county level. Last year the figures were 69 and 35,000. The Central Committee that closed its plenum on today (29/10/26) seated 279 of the 376 members and alternates it began with in 2022, and confirmed 33 more expulsions, among them Zhang Youxia, the country’s top general until his arrest this year. The campaign now reaches the former aides of Wang Qishan, the man who ran its first phase.

If the remedy worked, the numbers would fall. However, they rise every year, and the complaint grows louder.

The Party has its own name for what is happening at the bottom. It calls them “lying-flat cadres,” borrowing the phrase ordinary Chinese use for giving up on a calculation that no longer works. Cities hand out “snail awards” to the slowest offices, and one county in Jiangsu urged staff to “find the lying-flat cadres around you.” The Party’s own newspaper has described the type: officials who “act cooperative on the surface” but have “no desire for promotion or advancement, and no one can do anything with them.” The discipline commission now counts the problem directly. In 2024 it handled nearly 94,000 cases of inaction and related failures, and inaction-type offenses made up seven of every eight bureaucratic cases it processed.

Chinese officials learned this lesson long before Xi. In the third century the essayist Li Kang wrote that when a tree rises above the forest, the wind will break it. Every schoolchild knows the line, and plenty of county secretaries still live by it.

Economist Albert Hirschman named the mechanism in 1970. Members of a failing organization have two responses: exit, by leaving, or voice, by complaining from inside. Loyalty is what keeps people using voice instead of walking out. Xi demands loyalty while making voice fatal, since an official with reservations is, by Xi’s own definition, two-faced. Formal exit is closed too, since an official cannot quit the Party and keep his career. With both doors shut, exit doesn’t disappear; it goes underground. The official stays on the payroll, files every form on time and withdraws the one thing no inspector can audit, which is effort.

Some scholars read lying flat as voice rather than exit, a quiet protest. It can be both: when open speech costs too much, withdrawal becomes the only speech left.

I spent forty years dealing with state-controlled counterparties across Latin America and Asia. The paperwork was always complete. The officials who had learned to fear a mistake more than a delay delivered on their own schedule, and no inspector ever wrote them up, because nothing on paper was wrong.

There is an honest limit here. In 1993 Hirschman revisited his theory to explain the fall of East Germany. In 1989, exit and voice worked together: Hungary opened its border, tens of thousands left, and the crowds who stayed behind in Leipzig began chanting “We are staying here.” Every train west announced that the state could no longer hold its people, and staying put became a demand for change. China has no Hungarian border. Lying flat produces not collapse but decay, a state that keeps its shape while losing its capacity.

That changes what to watch. The plenum communiqué reported unity, and it was accurate, because fear buys consensus in the hall. It promised to “make full use” of macroeconomic policy and put no number on it. The test is elsewhere: whether local governments, starved of land-sale revenue since the housing crash, actually spend the stimulus Beijing announces, when every project they sign becomes a file an inspector can open later. A quiet bureaucracy is not a capable one. In June I argued that the discipline apparatus grows whether or not the problem improves. This is what the growth costs.

Xi has a favorite line for his campaign, borrowed from a Tang poet who spent twenty-three years in exile:

“Beside the sunken ship a thousand sails pass; before the sick tree, ten thousand trees come into spring.”

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

Growth: fast, but narrow

August’s pace is partly a base effect: the same month last year grew only 1.61%. Still, the smoother trendcycle measure rose 7.11%, and activity from January through August is up 6.07%, against 4.24% a year earlier. The quarterly accounts agree. GDP grew 6.4% in the second quarter, the fastest since the third quarter of 2023, when it reached 7.9%.

The growth comes from a few sectors. Transport and storage led the second quarter at 12.8% and commerce followed at 11.6%, while construction grew just 1.3% and information and communication shrank 1.8%. INEC credited commerce, logistics and tourism again in August and noted that construction lost momentum.

The platform: Canal, ports and the Free Zone

The Canal closed fiscal 2026 with 13,951 transits, up from 13,404, and tonnage rose 6.5% to 521 million CP/SUAB tons. Preliminary, unaudited revenue reached $6.6 billion, 16.6% more than the year before.

September transits fell 1.8% from a year earlier, a small dip worth watching with El Niño forecast for 2027.

Ports moved 7,605,946 TEU through September, 3.2% more than a year earlier, with Manzanillo’s 18.7% gain offsetting losses at Balboa and Cristóbal during their change of operator. The Colón Free Zone is the standout: re-exports rose 26.5% from January to July and 35.9% in July alone. Venezuela bought $256.2 million more than a year earlier, Costa Rica $204.7 million more and the United States $167.9 million more.

Prices and jobs

Consumer prices rose 0.5% in September after 0.2% in August, lifting annual inflation to 2.8%, the highest since July 2022. Fuel did most of the work: the transport component is up 11.2% on the year despite subsidies.

Unemployment fell to 8.5% in August from 10.4% in September 2025, with 77,231 more people working and informality down to 45.8% from 47.1%. The detail is less flattering. The share of workers in full employment fell from 55.4% to 46.2%, and the number working full hours for less than the minimum wage rose from 213,370 to 413,155.

Public finances

The non-financial public sector ran a deficit of $3,387 million from January to August, $560.7 million less than a year earlier and equal to 3.56% of GDP, just over the 3.5% ceiling for the full year. Revenue grew, but current revenue still came in $958 million below budget, and interest payments rose 6.1% to $2,084.8 million. The ministry is counting on year-end Canal contributions and state-company dividends to land inside the ceiling, and says it will not use the fiscal waiver to pay for earthquake reconstruction.

What to watch

The October 9 earthquake. A magnitude 7.7 quake on the Azuero Peninsula damaged schools, health centers, bridges and homes in the interior. The USGS’s preliminary loss model put possible losses between 1% and 10% of GDP, a model of exposure rather than a count of damage. None of it shows in the data yet.

El Niño and the Canal. The Canal expects fewer transits and less revenue in 2027 as drought returns.

Next releases. INEC publishes the IMAE about 51 days after the month ends, so September’s reading should arrive in mid-November alongside October prices. Thirdquarter GDP follows in December.

The read

Look past the headline at who is doing the growing. Transport rose 12.8% in the second quarter and commerce 11.6%, while construction managed 1.3%. The sectors that sell to strangers abroad, ships through the Canal, boxes through Manzanillo, re-exports from Colón to Caracas and San José, are pulling the economy forward, and the sectors that wait on permits and public budgets are not. The labor survey tells the same story from the other side: 77,231 more people at work, yet the share in full employment fell from 55.4% to 46.2%. Growth that arrives through the ports does not reach the payroll on its own.

Logistics and transport are up and the unemployment rate is trending down, but growth earned abroad will not show up in Panamanian paychecks by itself. Keep an eye on construction.

Figures as reported by INEC (Contraloría General), the Ministry of Economy and Finance, the Panama Canal Authority, the Panama Maritime Authority and the Colón Free Zone, via the outlets linked in Sources; each row there carries its link. Canal revenue is preliminary and unaudited. Written with Claude and reviewed before publication. Corrections welcome. Not investment advice.

• INEC IMAE, August (Infobae)

• Q2 GDP (Infobae)

• Q2 GDP by sector (MEF)

• September CPI (La Estrella)

• Labor survey, August (El Capital Financiero)

• Canal fiscal 2026 results (La Prensa)

• Port traffic through September (Infobae)

• Free Zone re-exports (Revista EyN)

• Fiscal balance through August (La Estrella)

• USGS earthquake loss estimate (La Estrella)

Norm Singleton of the Market Institute made the strongest version of the industry’s case this month in RealClearMarkets. His answer to rising electric bills is a set of promises: Microsoft and Anthropic have pledged to cover any cost increase their data centers cause, and other firms have signed the White House Ratepayer Protection Pledge. He also hopes someone will revive the underwater data centers Microsoft abandoned. China already has, off Hainan and Shanghai, but seawater cools servers. It doesn’t build power lines.

The pledges were signed at the White House in March. The grid operator for Virginia and 12 other states buys power years in advance, at auction. This summer its own watchdog found that data centers drove 38 cents of every dollar in that auction. Moody’s, which prices risk for bondholders and has no stake in fairness, explained why: the system “socializes new build costs across all customers.” A pledge doesn’t change that formula. It sits on top of it.

For two decades the arrangement held because the cost was spread thin, and a few dollars on a monthly bill doesn’t fill a hearing room. Economist Mancur Olson described the pattern sixty years ago: the many who each pay a little stay home, while the few who gain a lot show up.

But costs don’t stay spread thin. A power line has to land somewhere. Dominion’s Golden-to-Mars line will run nine miles through eastern Loudoun, and homeowners there are fighting a ruling that lets it cross their property. Substations need land, and the land has owners. Once the cost had a street address, the people living there filled the hearing room, and the board voted with them.

On October 20, the same day the final vote slipped, Amazon learned what that room costs. It had paid George Washington University $427 million for its old Ashburn campus, planning four data center buildings. Now it will withdraw the plan, give the county 25 acres worth about $65 million, put $5 million into the housing fund, and drop any lawsuit over what the land could hold. That is a price, paid to the county whose residents could see the site. Nobody at that table priced the next power line.

I owe my August piece a correction. The water scare was still a scare, and the tax revenue was still real. But the substation was real too, and I left that piece of it out.

The fix is already running in Ohio. Since 2025, AEP Ohio has required large new data centers to pay for at least 85 percent of the power they reserve, whether they use it or not. Once that rule was on the bill, the utility’s projected new demand fell from 30,000 megawatts to about 5,700. Most of the line was speculation, and a price sorted it out in a way no promise could. The industry’s trade group opposed the rule while insisting it is “committed to paying its full cost of service.” Manufacturers have taken it to the Ohio Supreme Court, arguing it singles out one industry. A pledge is a promise. A price is a mechanism. At the end of August I argued these deals needed standard terms. This is what standard terms look like.

In the Free Zone, nobody builds a warehouse to a tenant’s specifications merely on a handshake. The tenant signs a lease for the space before the first column goes up. That isn’t hostility to tenants. It is what lets the next warehouse get built.

The pledge is a sheet of paper signed in Washington. The power line is nine miles of towers running through eastern Loudoun, and its cost gets tacked on to the electric bill.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

In 1914, Felix Frankfurter began teaching at Harvard Law School. For years he chose the clerks who sat beside Justices Holmes and Brandeis, and his students would go on to staff the New Deal. In 1939, Franklin Roosevelt put the professor himself on the Supreme Court. Over twenty-five years, one man had climbed every rung.

At Howard, Charles Hamilton Houston built a different ladder, one that ran through the courtroom rather than the bench. His student Thurgood Marshall won Brown v. Board of Education in 1954 as an advocate and joined the Court in 1967. Brown has a serious defense in the Fourteenth Amendment’s own words, whatever the Court said at the time.

Frankfurter’s is the ladder Ivette Alé-Ferlito pointed to at the end of Part 1 of this series, when she said La Defensa’s three Los Angeles judges were clearing a path to the Supreme Court. It has been climbed before, even from an elected trial court. In 1974, Arizona voters put Sandra Day O’Connor on the Maricopa County Superior Court. Seven years later she was on the Supreme Court.

Historian Paul Moreno’s review of Supremacy, a new book by Harvard law professors Nikolas Bowie and Daphna Renan, explains why the story has two means to an end. Progressives facing a hostile court have always had two answers. One attacks the court’s power: strip its jurisdiction, deny that judicial review was ever legitimate. The other changes who sits on the bench, through law schools and clerkships. Roosevelt’s court-packing plan failed in 1937, but deaths and retirements gave him seven of nine justices anyway.

For a generation, the second strategy worked. Liberals won at the Court, and they stopped complaining about its power. However, as the Court drifted right, the old complaint came back. Bowie and Renan now argue that the Court has taken more power than the Constitution gives it, and that Congress should cut it back by instituting term limits and expanding the number of judges. Moreno calls this buyer’s remorse: liberals embraced a powerful Court after Brown and regret it now that the other side holds it. La Defensa is putting its people on the bench from the bottom. The professors want to shrink what the bench can do at the top.

The two strategies look like opposites. Economist Thomas Sowell would say they share a premise. In A Conflict of Visions, he traces two ways of seeing the world, and they produce two kinds of judges. The constitutionalist asks whether the court applied the rules as written, regardless of who wins. The perfector asks whether the court made society more just. Sowell grants that serious people can hold both views in good faith. But a perfector values a court for what it delivers, so he staffs it while he can and weakens it once he can’t.

In Philadelphia, which has nothing to do with La Defensa, the district attorney’s office has repeatedly conceded, after trial, that convictions should be thrown out. Over two years, three courts found those concessions dishonest. As a fix, the Pennsylvania Supreme Court now requires every trial judge in the city to bring in the state Attorney General to verify any such concession.

In September, federal Judge Paul Diamond referred District Attorney Larry Krasner and a deputy to the Justice Department, after four of the office’s own supervisors testified they had been urged to mislead the court. On October 14, he refused to step aside. Krasner disputes their account, and nothing has been charged. In Diamond’s words, “There are few attorney transgressions more serious than lying to the Court.”

Consider the scene when a prosecutor concedes. The defense agrees, because the concession is what it wanted. The victim has no lawyer, and neither do the neighbors. In The Experiment Nobody Evaluated, I wrote about Sandtown-Winchester, once Baltimore’s most violent neighborhood, which went a full year without a killing. The people on blocks like those pay for a false concession, and they are never represented in the room. One person is left to ask whether the concession is true, and that person wears a robe. The Supreme Court said so in 1942: a prosecutor’s confession of error “does not relieve this Court of the performance of the judicial function.”

Now move one rung up the ladder. Seat a judge who sees the bench the way La Defensa’s founders describe it, as one step toward ending incarceration, and who measures a concession by where it leads rather than whether it is true. Prosecutor, defense and bench are now aligned. The conviction is vacated, the prisoner walks out, and the record shows three officers of the court in agreement.

The right should not read this comfortably. The Federalist Society climbed the same ladder, and the president who filled three seats from its list described the result in a perfector’s terms: in 2023, Donald Trump boasted that he “was able to kill Roe v. Wade.” A conservative who prizes the Roberts Court for its outcomes rather than its method is a perfector too. His buyer’s remorse arrives the day the ladder changes hands.

The defense worth building is a second check, one that does not depend on any single officer of the court getting it right. In Philadelphia today, when a prosecutor concedes, a lawyer from the Attorney General’s office walks into the courtroom. He does not ask whether the concession is wise. He asks whether it is true.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

Two articles landed in my feed on the same morning. One, in City Journal, was about three trial judges in Los Angeles. The other, a Law & Liberty review of a new book by two Harvard law professors, was about the Supreme Court. Neither mentions the other. Read together, they describe one project working from both ends of the American judiciary.

Start where the City Journal piece starts, in Courtroom 3 of the Marin County Hall of Justice, August 1970. A seventeen-year-old pulled a gun from under his coat, armed three inmates with weapons Angela Davis had bought, and took the judge, the prosecutor, and three jurors hostage. Within the hour the judge was dead, along with three of the attackers. I traced Davis’s intellectual formation in The Man Who Taught Angela Davis to Stop Waiting for the Workers. The point here is simpler. The attempt to break a court by force failed.

The organization at the center of the City Journal story has chosen a different method. La Defensa was founded in Los Angeles in 2020 by Eunisses Hernandez, now a city council member, and Ivette Alé-Ferlito. Both want to end incarceration. Alé-Ferlito has written that electing sympathetic judges is a harm-reduction step inside a broader abolitionist strategy, useful only as part of the larger goal.

What follows is how a small group wins seats on the largest trial court in the country. Each step is ordinary on its own.

It begins with a pipeline. La Defensa runs a judicial leadership academy that teaches public defenders how to campaign, raise money, and reach voters. Ericka Wiley, a deputy public defender for more than two decades, came through it.

Then come a few large checks. The group’s Justice PAC, a project of Tides Advocacy, drew support from Liz Simons, heiress to the Renaissance Technologies fortune, and Patty Quillin, wife of Netflix co-founder Reed Hastings. From 2022 through 2024 it spent at least $910,000 on judicial races.

The third step is the one that makes the rest work. La Defensa did not have to persuade millions of voters, because almost nobody researches a Superior Court race. Voters use shortcuts: the party slate card, the union recommendation. George Turner carried the endorsements of the Los Angeles County Democratic Party, the public defenders’ union, the County Federation of Labor, and the local chapter of the Democratic Socialists of America. Win the handful of committees whose cards voters carry, and the voters follow.

Political scientist James Q. Wilson called this arrangement client politics: benefits concentrated on a few people who care intensely, costs spread across many who barely notice. The few organize because the stakes are large for each of them. The many stay home because no single judicial seat is worth an evening of their time. Wilson also spent his career on crime, and he would have recognized the catch. The costs are not diffuse where they land. As I argued in The Experiment Nobody Evaluated, a small number of repeat offenders commit a large share of the violence, and their neighbors pay for every release. The cost is concentrated on particular blocks and dispersed only in the politics, among people who write no checks and attend no endorsement meetings.

The results so far are modest, and honesty requires saying so. Over two cycles La Defensa backed seven candidates. Three won: Holly Hancock in 2022, Turner and Wiley in 2024. Four lost. The three winners sit on children’s, family, small claims, and eviction calendars, not felony trials. The Los Angeles Superior Court has some 500 judges in nearly 600 courtrooms. Three seats is a rounding error.

It is also a foundation. Most California trial judges reach the bench by gubernatorial appointment, after vetting. Open-seat elections are the side door, and La Defensa walked through it. Once inside, a judge serves six years, and an incumbent with no challenger is re-elected without her name appearing on the ballot. A seat won once is usually held for a career. A seat lost costs one cycle, and the academy can train another class.

Voters can correct a district attorney. Los Angeles did, removing George Gascón in 2024 with 61.5 percent voting for his challenger. A district attorney is one name with one record, and voters could find him. Nobody can find five hundred judges.

None of this is unique to one side. The Federalist Society spent four decades building a pipeline of judges from the right, and its founders would recognize every step La Defensa has taken. Organizing to put your people on the bench is how American legal movements have always worked. What distinguishes La Defensa is what its founders say the bench is for.

The Latin adage says the drop hollows the stone, not by force but by falling often. In 1970, the radicals in Courtroom 3 tried force, and it failed within the hour. Their heirs have chosen the drop: a seat, then another, each term renewed in silence, the academy always recruiting.

Alé-Ferlito has said where the drops are meant to fall. “We’re actually clearing the path for judicial transformation, all the way up to the Supreme Court.”

Coming in Part 2: The ladder she describes has been climbed twice before.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

The instinctive response is a border response: find the men, check the papers, watch who comes home from Kyiv. Washington has moved that way, and the State Department says it works with Ukrainian counterparts to limit the transfer of battlefield skills. That response is right, as far as it goes. Every forged passport caught in Kyiv makes the next one costlier, and a cartel that pays more for each instructor will buy fewer of them. Screening works. The real question is for how long.

A US assessment reported in September contains the detail that answers the question. According to a Mexican intelligence official, Colombian drone specialists returning from Ukraine are offered large sums, and sometimes flown to Mexico, to train groups of droneros, drone pilots. The student who crossed the ocean is no longer the unit that matters. He is a teacher now, and his classroom is in western Mexico.

A skill learned at the front does go stale. Ukrainian pilots relearn their trade every few weeks as Russian jamming changes, and a man who left in 2025 teaches 2025. Against a Russian brigade, that lag would be fatal. Against a state police convoy or a rival cartel with no electronic warfare at all, it costs almost nothing. The cartels do not need the frontier, only the basics, taught well, to enough people. Once the first teacher has trained ten students and three of them can teach, the checkpoint is guarding an empty road.

Economist William Baumol saw the deeper pattern across a thousand years of history. In a 1990 paper he argued that the supply of entrepreneurial talent changes little from one society to the next. What changes is the payoff. In medieval Europe, the man with a gift for organization and risk often went to war, because ransom and plunder paid better than building mills. It was the same kind of talent that later dug canals and financed railways, and the rules of the game decided which. Flying a drone is a learned skill rather than a gift, but that sharpens Baumol’s point. A gift is scarce. A skill that can be taught in a field, with parts bought online, becomes cheap, and whoever pays for a cheap skill decides where it goes.

Ukraine is the clean demonstration of the productive half. As I wrote in “Big Defense Just Lost Its Alibi,” Kyiv’s e-points system pays units for verified results and lets them spend those points in a marketplace of competing drone makers. That rule pulled programmers, hobbyists and machinists into a defense industry that iterates in weeks rather than years. Move the same hands to a country where the best-paying buyer is a cartel, and the same skill comes out the other side.

None of this is Ukrainian policy, and the distinction matters. Russian channels have spent the year claiming Kyiv trains the Jalisco cartel. The record shows something more ordinary: forged documents slipping through lax vetting in a foreign legion recruiting under wartime pressure. That is a screening failure, not a program.

This should make those of us who trust markets uncomfortable. A price system does not care what it allocates. It found the droneros the way it finds crane operators and coffee pickers: quickly, without a plan, at the going rate. Whether that is good news depends entirely on who is paying, and on what the rules make expensive.

That puts the lever in Mexico, and I know how that sounds. Independent estimates put impunity for homicide there above 90 percent, and “prosecute better” has been the advice for thirty years. But a drone unit is easier to convict than a sicario. It buys motors, batteries, frames and release mechanisms, leaves purchase records and shipping manifests behind, and trains in open country. Prosecutors who follow the parts rather than the bodies have a paper trail most cartel work never leaves. That raises the first cost, the odds that cartel work ends in a cell.

The second cost is what a legal employer offers for the same hands. A drone trade that is cheap to enter, in crop spraying, survey work and infrastructure inspection, will not outbid a cartel for its best pilots, and it will train more pilots than Mexico has today. That is the honest objection. The answer is that the cartels are training pilots anyway. What a legal trade changes is the choice facing a young man in Michoacán who has not been recruited yet: a licensed job at a modest wage, or a patch and a short life. The cartel can buy a few hundred aces. It has to make the rest, and a legal trade raises the price of making them. Screening buys time. The wage decides what the time is worth.

The Jalisco cartel’s drone units reportedly train in squads of ten pilots with twenty drones between them, and some wear patches that read Operadores Droneros. A patch is how a trade announces itself. Somewhere in western Mexico, a Colombian who learned under Russian artillery is standing in a field with his ten students, and the one who learns fastest will be teaching his own squad by spring.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

The problem was never the data. It was the tool.

I wrote in April about how Baltimore accidentally ran the control group on the prosecutorial side of this argument. Charles Fain Lehman documented in City Journal this past September how progressive prosecutors built careers on the premise that the system’s problem was over-prosecution of repeat offenders. Philadelphia’s homicides rose 64 percent under Larry Krasner. Baltimore’s fell 60 percent when Ivan Bates reversed course and began sentencing the repeat violent offenders the data had already identified. Lehman was right about the prosecutorial failure. But the prosecutorial failure is one expression of a deeper error the whole system keeps making. Policing itself made the same mistake first, corrected it once, and is making it again right now.

In 1994, Bill Bratton brought CompStat to the NYPD. The system mapped where crime happened. Precinct commanders were held accountable for patterns in their territory. New York’s homicide count went from 1,946 in 1993 to 352 in 2015. The data insight was real: crime clusters geographically. The tool worked.

Then the system simplified the insight. If crime clusters in certain places, enforce hard in those places. By 2011, the NYPD was conducting 685,000 stops a year. A federal court found the practice unconstitutional on racial profiling grounds: the geography had become a proxy for race, and the Equal Protection Clause does not permit that substitution. A generation of young men in certain neighborhoods had been stopped, searched, and released with nothing found.

CompStat mapped where crime happened. That was true and useful. But a map of where crime happens tells you where to send police, not who to watch. Stop-and-frisk was what happens when you confuse the map for the territory: you police the geography, which means you police everyone in it. The data was always saying something more specific. A few dozen people in any given district were generating most of the violence. The map could not show you their faces. Precision policing is what happens when you stop reading the map and start reading the list.

When Bratton returned to the NYPD in 2014, he rebuilt CompStat around individuals rather than geography. Stops dropped by more than an order of magnitude. Crime kept falling.

Baltimore followed the same logic. Researchers at the University of Pennsylvania analyzed who was generating the city’s violence: across every district, under 2 percent of residents, older men with extensive criminal histories, appearing repeatedly as both victims and suspects. The city built its response around that group, not around the neighborhoods where the violence occurred. Between 2022 and 2025, homicides fell 60 percent. A 2026 paper from the National Bureau of Economic Research confirmed the strategy drove the reduction. Sandtown-Winchester, historically the most violent neighborhood in the city, went a full year without a killing.

The lesson was hard-won and specific. The tool is not the map. The tool is the list.

On October 2nd, Senator Bernie Sanders et al, introduced legislation to ban federal agencies from using automated license plate readers. Two days before that, a federal judge in Oklahoma ruled that a Tulsa sheriff’s deputy violated the Fourth Amendment when he used Flock Safety’s network of roughly 120,000 surveillance cameras to pull a month of a driver’s travel history without a warrant. The reason for the search: the vehicle had a California plate. Judge Sara Hill called it “indiscriminate mass surveillance.” The word had been used before. In 2013, a federal court used the same word to describe stop-and-frisk.

Flock made vehicle movement legible. Every plate, every camera, permanent archive. The state could see where anyone had been. CNN reporting identified at least 50 officers accused of misusing the system, including 26 who used it to track current or former romantic partners. The hardware was different. The failure mode was the original one: the schema is tractable, the judgment is hard, so the system used the schema.

James C. Scott spent a career documenting how states build tools that make complex local knowledge legible, then mistake the simplified category for the thing it was meant to represent. Neither a geographic hot spot nor a database of every vehicle movement in the country tells you which specific people are generating the harm. The harder work is building the list, then enforcing it.

Baltimore built the list. A federal court in Tulsa confirmed what happens when you don’t. The system searched the highway for the California plate rather than the person the data had already identified.

The schema keeps changing. The mistake keeps arriving in the same shape.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

The caller was the best foreign customer the American arms industry has. Saudi Arabia carried $129 billion in active sales cases before the $142 billion agreement announced in Riyadh in May 2025, and the orders kept arriving while the Crown Prince waited on the line: nearly ten thousand guided bombs notified to Congress on September 4, and 48 F-35s approved on September 17. The kingdom had asked for airstrikes and received a data feed.

On the ground, the order book bought very little. The Houthis took Mocha on September 10, and within days they held the Yemeni shore of the strait down to Perim island in its mouth. American intelligence and targeting help did not change the result, and the kingdom could not stop the advance. Regional officials say it is running low on interceptors and has gone to France, Britain, Pakistan and Egypt for air defense. Each American interceptor costs about $4 million and takes more than two years to build, against drones that cost as little as $20,000, the arithmetic I traced in “Twenty Thousand Dollars.” The order book cannot hurry the factory.

The kingdom never meant to fight alone. For decades it kept its army small and split among rival commands, so no general could turn it on the palace, and it filled the gaps with foreign contractors and trainers. Riyadh also knew what the order book did not buy. It spent years pressing Washington for a binding defense treaty, and last November it settled for a strategic defense agreement and the title of major non-NATO ally, which under American law carries no security commitment. The purchases were premiums on a policy Washington never issued, and in September Riyadh filed a claim anyway.

Set that claim beside Israel. Between 2020 and 2024, Saudi Arabia took 12 percent of American arms exports and Israel about 3 percent, according to SIPRI. Most of Israel’s purchases, though, are paid with American military aid, roughly $3.3 billion a year in grants, and American law requires that no Saudi sale erode Israel’s military edge. The smaller buyer, whose weapons Washington finances, fights its own wars, and in this one it fought beside Washington rather than asking Washington to fight for it. The largest buyer, paying cash, got its call returned with targeting data.

An arms sale binds the buyer, not the seller. A country that buys American jets needs American parts and munitions for thirty years, while America’s obligation ends at delivery. Aid reverses the direction. When Washington pays for a country’s weapons, Washington has money of its own at stake. Riyadh paid its own bills and ended up dependent on Washington. Washington pays Israel’s and has ended up invested in Israel.

Why Trump said “no” is a separate question, and the record allows two possible answers. One is that Riyadh was paid in its own coin. In January, a senior Gulf official said the kingdom would keep its airspace closed to any strike on Iran. In May, two American officials said Riyadh refused Prince Sultan Air Base and its skies for Project Freedom, the escort operation meant to break Iran’s hold on Hormuz, a claim a Saudi source denied. In August the kingdom signed a mutual defense pact with Turkey and Pakistan, neither of them a friend of Israel. A kingdom that closes its skies to Washington’s war can hardly be surprised when Washington’s war stays out of its sea. The other answer is the one Trump’s officials give: a second front in Yemen would pull forces from the war that matters, and after the interceptor arithmetic above, it is a defensible call. Under either answer, they attached a condition that rewards a slow reading. No American strikes, so long as the Houthis leave non-Saudi shipping alone.

The new pact produced a courier. When missiles fell on Saudi cities, Pakistani officials said their country would not enter Yemen, and Islamabad’s contribution was to carry a message to Tehran asking it to restrain the Houthis.

Washington does fight for the straits it counts as its own. Navy SEALs spent four months clearing mines from Hormuz, divers working at night to fix charges to each mine by hand, and the oil has come back, as much as 90 percent of prewar barrels by Washington’s count. At the Bab al-Mandab, Washington will act only if the Houthis hit someone else’s ships.

In January, the refusal ran the other way. That senior Gulf official gave Fox News the kingdom’s reason for keeping American jets out of its skies: Saudi Arabia would not lend its airspace to “a war Saudi Arabia is not a part of.”

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

The essay is being read as a confession about race and school choice. It’s actually a confession about something with nothing to do with either, a question worth asking about any institution a parent trusts with a kid, regardless of who runs it: what happens when a school is allowed to grade its own homework.

For nine years, every signal coming from inside that building said Hannah-Jones’s daughter was fine. Report cards said fine. Honors placement said fine. A teacher’s praise said fine. Every one of those signals originated with the institution being asked to certify its own performance, and at different points every one of them was false. A literacy coach eventually discovered the teacher had likely never run a single one of the reading assessments the state required. A folder of ungraded math homework, opened at the family’s dining table one December night, showed a daughter who believed she was mastering fractions she’d never actually learned. This isn’t a story about one bad school. It’s what happens inside any institution that faces no outside check on its own numbers: not fraud, most of the time, just a slow drift toward believing its own paperwork, because nothing costs it anything for doing so.

Nobody campaigns on abolishing accountability. They campaign on abolishing the specific number that would prove them wrong. The test becomes biased, the exam becomes inequitable, the screener becomes deficit thinking, and the argument can even be right in some of its particulars. But opposing the instrument and opposing the underlying fact produce the identical result: an institution nobody outside it can audit. Hannah-Jones’s daughter’s school never announced it wasn’t going to check. It just kept telling her, through every clean report card, that no checking was needed.

What actually caught it, twice, was a number the school never got to write. The first time, a literacy coach ran the assessment the classroom teacher had been skipping. The second time, four years later at a demanding public high school, a state algebra exam Hannah-Jones’s daughter had been sure she’d aced came back failed, then failed again, until a tutoring center traced it to gaps as far back as long division. Both corrections arrived from outside the building making the claim.

Maud Maron made a version of this argument years before that algebra exam ever came back. She was a twenty-year Legal Aid Society public defender, a Democrat at the time, and an elected member of the local school board. At the height of the 1619 Project, Hannah-Jones’s own Pulitzer-winning reframing of American history around slavery’s legacy, Maron spent that cultural moment defending the single standardized admissions test for New York’s specialized high schools, the same design Hannah-Jones herself was calling racist. It cost her something real: in 2020, after she published an op-ed against the era’s race-based school politics, her own colleagues, the Black Attorneys of Legal Aid Caucus, put out a statement declaring “Maud is racist, and openly so,” offering no evidence for the charge. Maron’s argument was that the families with the least institutional pull, immigrant parents with no English, no connections, no way to charm a principal into a favor, had the most to lose the moment that number stopped being fixed and started being discretionary. Hannah-Jones’s own reporting ends up proving her point without meaning to: a standardized number is often the only thing a poor family has that a well-connected one can’t out-charm. Where Maron’s fight stays inside the politics of one admissions exam, the failure her essay documents is bigger than that. It’s every unaudited number a school hands a parent, not just the one guarding the door to eight schools.

For most of those nine years, Hannah-Jones tried the only tool available to a parent who’s decided to stay: complain, meet, escalate, write a check the school’s own PTA budget couldn’t match, wait. None of it worked, because none of it touched the actual defect. The numbers coming back from inside the building couldn’t be trusted no matter how hard or how generously she pushed on them. Her daughter eventually solved her own problem the only way left open to her: she picked a different school and walked out the door.

That instinct is scaling up nationally right now, under both parties. Nineteen states now let a family take its public education dollars to a private school or a homeschool curriculum, up from zero in 2021, the fastest reversal in American schooling in a generation. It isn’t a clean win, though, and it’s worth saying plainly why. A voucher only routes a family toward something better if the number it’s choosing on is itself real, and most of the ratings families lean on under these new programs come from the same unaudited self-report that failed Hannah-Jones’s daughter for nine years, just wearing different politics. What travels faster than any of that paperwork is what parents tell each other, which class is real and which principal has stopped answering emails, a channel no institution writes and none can spin. Hannah-Jones had that signal in hand and didn’t read it as a signal: when her daughter’s effective pre-K principal retired, nearly every other family with the means to leave did, inside a year. She names it in her own essay as loss. It was also the truest number she had.

Hannah-Jones closes her own essay with the line that should have been the whole argument: there is no absolution here. Not for me. Not for you. Not for any of us. She’s still writing as if the sin was picking the wrong side of a fight about race. It wasn’t. It was trusting a number nobody outside the building was ever allowed to check.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

This is at least the fourth wave of federal muscle into Sinaloa since September 2024, when the U.S. arrest of cartel co-founder Ismael “El Mayo” Zambada split the organization into two warring factions, Los Chapitos and La Mayiza. Two years into that war, the government has real numbers to show: 1,026 arrests, 2,031 firearms seized, nearly 48 metric tons of drugs confiscated, 672 clandestine labs shut down this year alone. Sinaloa’s own prosecutors report daily homicides down 26.1 percent statewide through August compared to the same period last year. By the state’s own metric, this is working.

And still Mazatlán ranks second nationally this year in homicides, feminicides, kidnappings, and rape. Still the government is sending another 1,800 men, federal men, because the layer of government actually stationed in Sinaloa year-round is the one Mexico City has the least reason to trust.

Mexican municipalities collect almost none of their own tax revenue. Roughly 80 to 90 percent of a typical municipal budget arrives as federal or state transfers, participaciones, already earmarked before a mayor sees a peso of it, which leaves nothing discretionary for something as basic as raising police pay on its own initiative. Sinaloa’s state police already earn the highest average salary of any state in Mexico, and that salary has climbed faster there than almost anywhere else since 2021. That isn’t an oversight. It’s a government visibly losing a wage war it can’t stop bidding on, because what a cartel is willing to pay for control of one plaza scales with what the plaza is worth to them, not with what a state budget can plausibly match. Mexico City’s experience shows the arithmetic plainly: police there average roughly 15,000 pesos a month against cartel recruitment offers running 18,000 to 20,000, and officers who take the offer don’t describe it as a complicated decision. As one report on the pattern put it, they won’t risk their lives for a quincena.

Here’s the piece the headline number leaves out. A rent, in the economist’s sense, is a payoff that exists because someone controls access to something scarce: a route, a plaza, a judge’s signature. Economist Gordon Tullock spent his career on the uncomfortable half of that idea: people don’t collect rents for free. They spend real resources defending them, resources that produce nothing, because the entire point of the spending is to keep a rival or a reformer from taking the rent away. That spending is the actual tax. Not the extortion itself. The war to keep it.

Mexico has actually tried a structural fix for this: a unified state police command, mando único, meant to pull policing out of 2,470 individually capturable municipal forces and consolidate it at a level where capture is, in theory, harder to arrange. It’s been proposed repeatedly since 2019 and stalled every time, because state governors don’t want to give up control of local police as a political asset. That’s a second rent, sitting on top of the first, and it explains why the fix keeps losing to the same incentive it’s supposed to correct.

The Chapitos and La Mayiza aren’t fighting over ideology. They’re fighting over control of the specific plazas, ports, and local offices that let a rent outlive the man who built it. Every commander they buy, every judge they intimidate, every rival they kill is Tullock spending, an outlay whose only return is defending an existing extraction racket against dismantlement, by a rival faction or by the state. A fight this well capitalized doesn’t register more troops as an existential threat. It registers them as a rounding error against the rent being defended, and a predictable one, since the troops are federal and temporary, and the plaza is local and permanent.

That’s why more men produce a homicide dip and a press conference without producing an end state. The state is competing on the wrong ledger, at the wrong level of government. It counts what it seizes. The cartel counts what it protects, and will always spend more to protect a going concern than a government on an election calendar, or a governor protecting his own turf, will spend to permanently close it down.

None of this argues for standing down. The 26 percent drop in Sinaloa’s homicide count means troops on the ground are saving identifiable lives this month. But saving lives this month and closing the rent are two different projects, and Mexico City keeps announcing the first while implying it has done the second.

Somewhere in Culiacán this week, a municipal police officer worked out, again, which side of plata o plomo he can survive saying no to. That ledger doesn’t show up in García Harfuch’s press conference. It never does.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

Tyler Cowen published a small model last week that gets at exactly this. An economy runs on two ingredients, he says. One is Intelligence: chess, math proofs, test scores, the stuff AI is suddenly very good at. The other is Polanyi knowledge, named for philosopher Michael Polanyi, meaning the unwritten, on-the-ground knowledge of how a specific office, market, or institution actually works. AI has flooded the world with the first ingredient. It has barely touched the second.

You can’t trade one for the other. Pour in more Intelligence and it doesn’t make up for a shortage of Polanyi knowledge, any more than a hundred extra keys help if you’re missing the one lock they fit. Why not just gather more data on Polanyi knowledge until it’s absorbed too? Because some of it isn’t sitting anywhere waiting to be collected. Polanyi’s own example was riding a bicycle: ask an expert to state the rule and you get something true and useless, turn the handlebars toward the fall. The knowledge lives in thousands of adjustments a body makes without ever converting them into a sentence. It’s produced fresh in the act itself and disappears the moment the act ends. There’s no file to find. That’s a very Hayekian idea: no planner, however brilliant, can outthink knowledge scattered across millions of people, in part because that knowledge was never sitting still long enough to be gathered up. A superintelligence facing the same wall isn’t a planner with better tools. It’s the same wall.

Economist Arnold Kling sharpened why the split falls where it does. He divides the world into three: digital, physical, and social. AI had a mountain of data to learn from in the digital world, so that’s where it got good first. The physical and social worlds never left that kind of record: we know the difference between catching an egg, a leaf, and a baseball without ever writing it down. Alpha School’s AI tutors can track exactly how far a student has gotten in a subject, but they don’t know what will make that student push harder today. A human guide supplies that. None of it shows up in a training set, no matter how the model was built, and no regulation changes that either way.

What makes this worth a second look is where else it turned up last week. A few days before Cowen’s piece ran, Dwarkesh Patel recorded a two-hour conversation with three AI researchers, from Zyphra, Thinking Machines, and Baseten, about recursive self-improvement, AI improving AI without a human in the loop, the actual mechanism any takeover scenario needs. None of them mentioned Cowen, Hayek, or Kling. They were talking shop, about training runs and data. And they kept landing on the same line anyway.

Charlie O’Neill drew it as a line between two kinds of work. Some tasks are cumulative: every discovery is permanent, a line you hold forever once you cross it. Other tasks are non-stationary: the ground keeps shifting. Being a lawyer’s assistant at a firm is non-stationary, he said, because you’re tracking a constantly changing web of relationships and unwritten habits. That’s Kling’s digital-versus-social divide, rediscovered in a meeting about training environments instead of an economics essay.

John Schulman, who helped build the technology behind ChatGPT, kept coming back to taste: good long-term judgment built slowly through real experience, not mountains of data. He also flagged the sim-to-real gap: labs can simulate math and code by the millions inside a data center, but they can’t simulate a model quietly learning how your office runs over six months, let alone how the world runs. Different words. Same wall.

Here’s where it gets interesting: the three don’t agree with each other. Economist Thomas Sowell described two ways of seeing the world. In the unconstrained vision, most obstacles are engineering problems, solvable with enough cleverness. In the constrained vision, some things can’t be substituted for, no matter how much of something else you throw at them.

Two of the three, O’Neill and Millidge, lean unconstrained. They think a fully capable AI worker, handling a month of complicated office work unsupervised, is one to three years out. Schulman leans constrained, stressing how hard taste is to train and guessing some fields will simply take longer because they lean so heavily on the tacit stuff.

None of them are guessing wildly. They’re looking at the same models, the same evidence. The disagreement isn’t about facts. It’s about how substitutable Cowen’s two ingredients really are, and which vision you hold quietly decides which side of both fights, doom and regulatory capture alike, you find yourself on.

That gives you something useful next time someone hands you a bold AI claim. Ask which vision it’s assuming. Does it assume tacit knowledge can eventually be trained into a machine, or regulated into safety, given enough compute or paperwork? Or does it assume that knowledge has to be lived, slowly, by someone who was actually there? The confidence attached to the warning is really just a stand-in for that answer.

Cowen ends his own piece noting that if someone, or something, tried to seize control of AI’s Intelligence sector alone, they’d have far less power than it looks like, because the other ingredient is missing. That’s the actual bound on Amodei’s warning, and on his critics’ fear of what a moat built from it could buy him. Three researchers spent two hours, without meaning to, proving Cowen right from the inside of the machine.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

On March 28, 1979, a valve stuck open at Three Mile Island’s Unit 2 reactor near Harrisburg, Pennsylvania. Cooling water drained. Part of the core melted. Radiation leaked, in amounts a presidential commission found too small to cause measurable health effects. Nobody died. Nobody has ever been shown to have gotten sick.

Twelve days earlier, The China Syndrome had opened in American theaters. Jane Fonda played a reporter uncovering a cover-up at a nuclear plant near meltdown; Jack Lemmon, the engineer who tries to stop it and gets destroyed for trying. The film’s tagline promised a catastrophe reaching the earth’s core. Michael Douglas produced it to convince Americans nuclear power was a lie the industry told them.

Then Three Mile Island happened, and the country already knew what it meant. Not because of the accident report, which took months and concluded the safety systems had largely worked. Because of the movie, which took two hours and had already told them the ending.

The fear outran the danger by a wide margin. Thornburgh’s actual advisory covered pregnant women and small children within five miles, some 12,000 people. An estimated 140,000 fled anyway, some from towns forty miles outside any evacuation zone, bank tellers describing customers emptying accounts to flee a disaster that, in the end, hurt no one. Most people today remember none of that. They remember a meltdown.

No new American reactor was ordered and completed for more than thirty years after that. Not because the accident proved the technology unsafe. It proved the opposite, even a stuck valve and partial meltdown killed no one. It proved instead that the public doesn’t wait for the accident report. It reaches for the script it already has.

Economist Bruce Yandle coined a name for this: Bootleggers and Baptists. His example: Prohibition-era Sunday liquor bans, backed by preachers wanting less drinking and bootleggers wanting less competition from Sunday sales. Neither needed to know the other existed. Both needed the same law. Watch for that pattern, not because it proves a conspiracy, but because it explains how a law gets built and stays built without anyone at the center coordinating it.

Watch it assemble again, around a different machine.

In September, researcher Jacob Coxon quit Anthropic and OpenAI within the same year, warning both companies were racing toward AI systems that could improve themselves faster than anyone could keep them aligned. Days later, Anthropic’s own alignment researcher agreed and put the odds of AI causing human extinction this decade above ten percent. Anthropic’s CEO followed with an essay calling for the industry to slow down, backed by a first step it can take alone: permanent, employee-level evaluators auditing its own models. The essay’s next step asks Washington for an antitrust waiver, letting rival labs agree on shared rules together, the opening move toward actual government coordination. Sam Altman and Elon Musk said yes within hours.

None of that needs to be fake to work as the Baptist half. Coxon has three years of pretraining work at OpenAI and nothing obvious to gain by leaving. His former colleague has equity in the company he just told the public not to fully trust. That kind of admission is expensive, and people don’t spend money they don’t have to.

But watch who else needed this law, for reasons that have nothing to do with that sincerity. A rule requiring permanent embedded evaluators is a line item for a company with Anthropic’s balance sheet. It is a wall for the startup two years behind it, and does nothing to Chinese models already free to download and run on a laptop with no company behind them, since there’s no front door left to post an evaluator once the software is loose in the world. Separately, and documented rather than suspected, a network of Chinese state media and CCP-linked funding has spent two years amplifying the American movement against AI infrastructure, not inventing objections, just turning up the volume on real ones. None of that requires either group to know the other exists.

That is the whole trick, worth sitting with even if you find the underlying fear credible. Ten percent, from someone with nothing to gain by saying it and everything to lose, is not a number serious people wave off. But a coalition doesn’t need a shared motive to produce a shared outcome, and the rule written under this kind of pressure tends to fit whoever helped write it, not the actual risk. That was as true of the reactor that never got built as it is of the model that hasn’t been trained yet.

The Unit 1 reactor at Three Mile Island kept running for another four decades after its neighbor melted down, generating power without incident until it closed in 2019 for reasons that had nothing to do with safety and everything to do with cost. It sat, the whole time, next to the concrete shell of Unit 2, and next to the outline of all the reactors that were never poured at all, killed by an accident that killed no one, and a movie that got there first.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

That single sentence does more to settle the “collapsing” argument than any tweet from last week. Reuters reported Houthi forces pushing government troops south toward Dhubab, the district that faces Perim Island across the strait’s narrowest channel, control of which has decided who holds Bab al-Mandab in every prior round of this fight going back a decade. Unconfirmed reports Thursday put Houthi fighters on the Hanish islands as well, though that one hasn’t firmed up yet and shouldn’t be treated as settled. What has firmed up: Saudi Arabia has answered with roughly 40 airstrikes, according to Houthi accounts, on top of the Houthi missile and drone assault on Saudi oil facilities earlier this week.

None of this reads like a group in collapse. It reads like a group that just took the one piece of coastline that turns “threatening the strait” into “holding the strait.”

Here’s where Hormuz becomes useful, not as backdrop but as a preview.

For four months, Navy SEAL divers, robotic boats, and sonar-equipped submersibles swept roughly 80 Iranian mines from the Strait of Hormuz, an operation CENTCOM’s Adm. Brad Cooper later described publicly. The Navy has since escorted nearly 1,500 commercial vessels through, at times moving dozens of tankers a day under armed protection. Trump declared the strait under US control. And yet shipping companies and maritime security advisers still believe only the areas near Oman are reliably clear, and remain uncertain about the rest, because as one tanker executive put it, it doesn’t take much to put a mine back given the volume of small-boat traffic crossing Hormuz every night. Four months of special forces work bought safer transit. It did not buy a normal insurance rate.

Economist Robert Higgs called this regime uncertainty: capital sits on the sidelines not when damage is certain, but when nobody can price what happens next. An underwriter doesn’t need proof of a live mine. He needs only to believe one could reappear, and the premium holds at wartime levels on a strait the US calls swept clean.

Now watch what just happened to the other gate.

The US spent four months and a covert special-forces campaign trying to squeeze the uncertainty out of Hormuz, and largely succeeded on the physical threat while failing to move the number that actually matters to a shipowner. Iran’s answer, whether by direct order or by an ally reading the room on its own, was to let the pressure show up somewhere else entirely. The Houthis needed no submarines, no mine-clearing fleet, no four-month campaign. They needed a port town and a coast road. The Houthis needed no submarines, no mine-clearing fleet, no four-month campaign. They needed a port town and a coast road. What took the US Navy a full covert war to almost pull off at Hormuz, the Houthis matched at Bab al-Mandab in a week.

That’s the mechanism worth naming: squeeze uncertainty out of one chokepoint and, if the underlying incentive survives, it reappears at the next one, cheaper to produce and just as expensive to insure against. Hormuz proves that even total physical control doesn’t reset a war-risk premium once. Bab al-Mandab is about to test whether a much smaller, much cheaper Houthi land grab can produce the same effect from scratch.

Nicholas Hopton, Britain’s former ambassador to Yemen, spent years dealing directly with Houthi representatives and calls them “an intractable and powerful Yemeni political and military actor.” That was true before Mocha fell. It’s harder to argue with now.

Somewhere in London this week, an underwriter is looking at two straits instead of one: the first swept by American divers and still uninsurable at a normal rate, the second just handed a port town by a force that was supposedly collapsing seven days ago. He isn’t asking who’s winning either war. He’s asking whether the whole corridor, both gates, just became one continuous risk instead of two separate ones.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

That's not a criticism of Anthropic specifically. It's the industry's standard shape now. The Wall Street Journal quantified it two weeks before this deal closed: nine of the largest tech companies carry roughly $3 trillion in AI-related commitments that don't appear as balance-sheet liabilities, against about $600 billion in disclosed trailing capex. None of this is hidden. The numbers sit in the companies' own audited SEC filings. The problem isn't concealment. It's that the accounting rule for when an obligation becomes a liability wasn't built for financing structures this large or this fragmented, so a real, multi-hundred-billion-dollar commitment can be fully disclosed in a footnote and still never touch the total-liabilities line that feeds a debt-to-equity ratio. Morgan Stanley's own accounting analysts flagged this on the record in April: these structures make it increasingly difficult for investors to assess companies' total potential leverage. Two separate problems live inside that gap.

The first is fragmentation. Nvidia holds the lease on the Nueces County building. Lambda operates it and sells the compute inside it to Anthropic. Hut 8 owns the physical structure and its own construction debt. Four parties, one $35 billion commitment, no filing that shows the whole number in one place. Meta ran a version of the same play with its Hyperion data center in Richland Parish, Louisiana: a fund controlled by Blue Owl Capital owns 80% of the joint venture that owns the building, an arrangement its own auditor, Ernst & Young, approved while flagging it internally as a "critical audit matter." That's Hayek's knowledge problem in its purest form: no single ledger was designed to hold the information a lender would need, so nobody's model can add it up even when every piece sits in public view.

The second problem is different, and this series has drawn part of it before. In August, this newsletter mapped a circle where Nvidia agreed to guarantee roughly $250 billion of OpenAI's lease on a SoftBank-linked data center: a company backing the customer of a firm partly owned by that customer's own investor, who was also the guarantor's investment. The mechanism repeats here. Nvidia's own balance sheet is doing work to manufacture demand for Nvidia's own product, and that demand gets reported back as organic growth. What's different is what Nvidia is holding. In August, it held a guarantee, a promise to cover someone else's exposure if the deal went bad. In Nueces County, it holds the lease itself, the underlying asset, not a backstop on it. The guarantor became the landlord.

This series opened on an even older version of the same precedent. Lucent and Nortel financed demand for their own telecom equipment in the late 1990s: Lucent extended roughly $8.1 billion in vendor loans to the carriers buying its gear, Nortel another $3.1 billion, and Nortel took a $16.1 billion loss in a single year when the demand it had financed failed to arrive, more than every dollar of profit the company had ever made. The dollar figures were smaller then, and everyone involved could still see the whole transaction on one ledger. Neither is true of the loop running through Nueces County.

None of this argues AI is overhyped. It may justify every dollar spent on it. That's a separate question from whether the financing can be measured by anyone pricing the risk. Railroads were the right technology in 1873, and pioneers still lost everything when Jay Cooke's bank collapsed under debt nobody could size up correctly. The railroads got built anyway, decades later, by people who paid nothing for Cooke's mistake. The value of what gets built and the soundness of what pays for it are not the same claim.

Pioneers are the ones who end up with arrows in their backs. The settlers who come later get the land.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

It’s a bad argument on its own terms. Economist Phil Magness pointed out the obvious hole in his review of Vance’s new book, Communion: Japanese strawberries don’t sell at American prices, the premium varieties run into the double digits per berry, and GDP already captures quality through the price itself. Vance mistook two differently priced products for identical ones and built a worldview on the mistake.

But underneath the bad argument is a real question, and it’s worth taking seriously even after you’ve thrown out how Vance answered it.

Some things resist being counted. A parent arguing with a toddler about carrots doesn’t show up in a national accounts ledger. Michael Polanyi, a chemist who became a philosopher of science, spent decades on exactly this problem. He called it tacit knowledge: the things we know but can’t fully say. A cyclist can’t write down the physics that keep him upright. A mother reading her child’s mood before he speaks a word can’t hand you the algorithm. Polanyi’s claim wasn’t that economics gets these things wrong. It’s that no formal system, economic or otherwise, can fully capture them, because they were never propositions to begin with.

Friedrich Hayek built a related, and connected, case decades earlier from the opposite direction. His problem wasn’t what can’t be said, it was what’s scattered too widely to collect. The knowledge that runs an economy, who wants what, who has what to spare, what a shortage in one town means for a surplus in another, doesn’t sit in any single mind or office. It’s distributed across millions of people, each holding a fragment. Prices are how that fragment moves without anyone having to write it down. This wasn’t parallel invention. Hayek explicitly absorbed Polanyi’s tacit knowledge concept into his own account over the years that followed, and the two men were friends and founding members of the same society in 1947.

That’s the standard Vance is reaching for when he says a strawberry disproves GDP. He’s right that something escapes the count. He is standing, whether he knows it or not, on ground Polanyi and Hayek surveyed with real rigor two generations before he was born.

He is also failing the test they set.

Polanyi and Hayek didn’t use the limits of formal knowledge as a reason to stop measuring. They used it as a reason to measure honestly, and to know exactly where measurement runs out. Vance skips that step. The Bureau of Economic Analysis has tracked the value of unpaid household labor, cooking, cleaning, childcare, for decades. Magness ran the numbers Vance never checked: from 1965 to 2020, adding household production to GDP drops the growth rate from 6.3 percent to 6.1 percent. The data Vance says doesn’t exist, exists, and it barely moves the needle he’s certain it should break.

The pattern repeats. Vance treats the trade deficit as important enough to justify a declared national emergency, then treats a soft producer price report as proof that economists don’t understand their own subject, when that report simply excludes the imported goods his tariffs touch. He reaches for GDP when it flatters the argument and declares it fake when it doesn’t.

Polanyi spent a career insisting that some knowledge can’t be forced into a spreadsheet. He never used that insight as a license to stop checking the numbers in front of him. Vance faced the same fork and took the other road.

Asked what he’d learned in Japan, he didn’t say he’d learned where measurement ends. He said, “Maybe economics is just fake.”

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

Nobel-winning economist Ronald Coase answered a version of this question in 1937, asking something that sounds too simple to need asking: why do firms exist at all, instead of everyone contracting with everyone else, transaction by transaction? His answer was that using the price mechanism has a cost of its own. Discovering the right price, negotiating terms, writing and enforcing an agreement, all of that is expensive, and a firm exists to economize on those costs by replacing repeated bespoke negotiation with standing internal rules. Anywhere a transaction keeps getting relitigated from scratch, expensively, in secret, and unevenly, that is usually a sign the transaction costs were never brought down, not that the deal itself is impossible.

That's how data center incentive deals keep getting made. Every negotiation starts over. Every term gets fought out behind a nondisclosure agreement. Every community discovers the shape of the deal only after the vote approving it. Read through Coase, the backlash looks less like opposition to data centers and more like the predictable cost of a transaction that never got standardized.

Take the rate class question first, because it decides everything else. Virginia’s fight over this came down to whether a hyperscaler that orders a given amount of power capacity should pay for it whether or not it uses all of it. Dominion Energy’s proposed rate class requires large-load customers to sign 14-year contracts covering the bulk of their contracted demand, used or not. Without a dedicated rate class, that cost gets amortized across every residential ratepayer in the territory, whether or not their street has a data center on it. A standard rate class turns that into a standing rule instead of a fight every utility commission has to have from scratch.

Then there is the exit nobody prices at the entry. North Carolina learned this with VinFast: a promise of 7,500 jobs, roughly 1,400 delivered, and a lawsuit in May to claw back the land and tens of millions in incentives. Capacity commitments that survive cancellation, and clawbacks that trigger automatically rather than requiring a state to go looking for leverage afterward, are standard contract terms doing the job standard contract terms are supposed to do: absorbing the cost of a bad outcome before it happens.

The secrecy is the part that costs the least to fix and does the most damage left as is. Power and water terms routinely stay locked behind an NDA until after the approval vote, so the people bearing the deal’s costs learn its terms only once reversing course is politically expensive. That opacity isn’t only friction either. It’s a bargaining asset: it lets a developer negotiate each jurisdiction cold, without one county’s terms setting a floor for the next one’s demands, and it lets local officials announce jobs and investment before anyone can organize against the terms underneath. Nobody’s fixed this on their own because the people writing the deal are the ones it currently benefits. A prominent politician once defended passing a bill before anyone could read it on the theory that the reading would happen afterward. Data center deals have quietly adopted the same sequence, minus the defense. Disclosure as a condition of the incentive, rather than a courtesy extended afterward, costs neither side anything the deal wasn’t already going to cost. It just moves the discovery of the price to before the negotiation closes, which is the entire Coasean point.

There’s a second reason disclosure matters right now, distinct from transaction costs. The same labs racing to build this compute are also facing pressure to hold back their most capable models from public release, for safety review, for security incidents, for reasons that don’t always get explained. When the deal-making that fuels that buildout happens behind an NDA too, it compounds the same problem: less of what these companies are actually doing stays visible to anyone outside them. A standard that puts terms on the record before the vote doesn’t fix opacity inside a lab. It at least keeps the negotiations that fund the lab’s growth from adding to it.

None of this requires treating hyperscalers as adversaries or communities as obstacles. It requires standardizing the negotiation the way any firm standardizes a transaction it expects to repeat. A jurisdiction offering a standard rate class, a standard clawback, and disclosure before the vote is not giving anything away. It is lowering the cost of doing the deal, for both sides, every time.

There is a reason to move now rather than treat this as optional. Every one of these fixes is a rule for a jurisdiction still choosing to negotiate. Orbital compute doesn’t lower the transaction cost of a data center deal. It moves the negotiation somewhere local governments can’t reach it at all, out of county commissions and utility boards and into a handful of launch licenses and federal agencies a continent away. No rate class, no clawback, no vote to disclose terms before, not because the deal-making disappears, but because it relocates to an address the states currently doing the relitigating will have no seat at. The states spending 2026 relearning the theory of the firm the hard way have a narrower window than they think to get the standard version of this deal built, before the leverage they’re fighting over changes hands entirely.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

It’s a compelling case, and Kelly is honest about how far we are from it. He lists the unknowns plainly: we don’t know the minimum energy intelligence requires, whether there’s a shared ingredient between a human brain and a silicon chip, or how you’d even measure “smart” on a single scale. He notes, correctly, that the brain runs on about 25 watts, a fact that should embarrass every data center under construction right now. His hope is that a small group of scientists, part neurobiologist, part physicist, part philosopher, will do for intelligence what Carnot did for heat: find the formula underneath the tinkering.

Hayek’s “fatal conceit,” laid out in his last book by that name, has already been drafted into service against this kind of ambition once. It’s become the standard rebuttal to AI regulators and central planners, the argument that no bureaucrat can out-compute a market by pretending to know what the price system already knows without anyone knowing it. That’s not what’s happening here. Kelly isn't proposing a regulator. He's proposing a scientist, someone to find the theory of intelligence he's after. The fatal conceit, if it applies at all, has to be aimed somewhere else.

Here’s where. Picture a market for tin. If the metal gets scarcer somewhere in the world, its price rises. That single number tells a smelter in Bolivia to dig more and a manufacturer in Ohio to use less, and neither one has to know why. This is the mechanism Hayek laid out in “The Use of Knowledge in Society”: the knowledge that made the price move, a strike, a war, a new deposit, stays scattered across a thousand people who never meet and never compare notes. Hayek’s point wasn’t that this knowledge is unknowable. It’s that it can’t be gathered into one place and written down as a rule. It only exists in the act of being used, distributed across the system, never centralized in it.

That’s a good description of a mind. Not “a computer that hasn’t been optimized yet,” but billions of interacting, context-bound processes, roughly ten times the population of Earth packed into a single skull, each one carrying a sliver of what the whole thing is doing, none of them holding the formula for the whole.

A price works because it’s a number that carries information without carrying the reasons behind it. Something like this shows up in the brain too. A connection between two brain cells doesn't come with a memo attached. It just gets stronger or weaker depending on how much it's used, a finding neuroscientists have called "neurons that fire together, wire together" since Donald Hebb's 1949 work, later confirmed at the cellular level as the connection physically strengthening with repeated use. The cells downstream don’t need to know why the path got worn in. They just use it. That’s the missing piece in the market comparison: not just that the knowledge is scattered, but that it’s scattered in a form other parts of the system can act on without reading the whole file.

The 25 watts Kelly is impressed by might be the tell. A system that cheap isn’t running an elegant unified algorithm underneath the mess. It’s running the mess, and the mess is the point, the same way a market’s efficiency comes from the fact that nobody is in charge of it.

If that’s closer to right, then the theory he wants isn’t merely undiscovered. It may be the wrong shape of answer for the whole question, the same way no one has ever written down a single formula for “the economy” and had it hold. What we have instead, and what has actually worked, is local theory: the tin price still gets explained by supply and demand at the margin, even though nobody explains the whole system that produces it. Carnot’s own formula only ever answered one question, the efficiency limit of an engine, not “heat” as such.

None of this means the search is wasted. It means the payoff looks different than Kelly is picturing. Not one formula that tells you in advance whether a training run is 1 percent or 90 percent of what’s physically possible, but a string of local, falsifiable, mechanistic claims, this circuit does this, this constraint produces that tradeoff, each one useful, none of them the grand unified theory of mind. That’s not a lesser prize. It’s the only kind of prize a spontaneous order was ever going to hand over. That's not a theory of intelligence. It's a theory of what a theory of intelligence can look like, and Hayek already wrote the only version of it that's ever held up.

Carnot's engine had a blueprint to reverse-engineer. A mind may have no blueprint at all, only billions of connections quietly strengthening or fading with use, no ledger anywhere that adds them all up.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

Cambridge and the degree-granting university both said the work had been checked and cleared. The professor, Jason Arday, had built a genuinely remarkable public story: nonverbal until 11, illiterate until 18, a Cambridge chair at 37, the youngest black professor in the university’s history. Times Higher Education reportedly had its own plagiarism story ready months earlier and shelved it after Arday’s lawyers sent a letter. Then Cofnas published. The story broke wide. Arday resigned. A week ago, he died.

This week, Ghent University suspended Cofnas. Not for getting the analysis wrong. Ghent’s statement is careful on this point: the action concerns his “public statements... regarding this matter,” not the underlying finding. Nobody has produced a rebuttal to the 188 sentences. The university moved against the man who counted them.

The French anthropologist René Girard spent a career studying exactly this pattern, and it is older than any university. He argued that human communities under mounting internal tension don’t resolve it by identifying the actual cause. They resolve it by converging, almost involuntarily, on a single figure whose expulsion restores the group’s peace. What made someone a viable target was never really their guilt. It was their capacity to absorb the group’s accumulated anxiety and be removed without the group turning on itself.

Girard’s less-quoted observation is the one that matters here. The mechanism does not require a transgressor. It only requires someone disrupting the group’s equilibrium. A person who names an uncomfortable, well-evidenced fact is disrupting the equilibrium just as surely as the original offender did. Structurally, the truth-teller and the transgressor can occupy the identical position in the crowd’s eyes.

Luke Burgis, whose new book carries mimetic theory into exactly this territory, made a point in a recent conversation with Tyler Cowen that complicates the tidy version of this story in a useful way. Asked whether a scapegoat still produces catharsis, Burgis said no. His reading is that the mechanism has stopped working cleanly, that Christianity itself exposed it too thoroughly for the old release valve to function, and that what we get instead is runaway violence: expulsions that fail to restore peace and simply generate the next target. That is a better description of this past month than the classical version. Arday’s death did not end anything. It escalated straight into Cofnas’s suspension, which has settled nothing either. The plagiarism finding stands unrebutted. The grievance has only multiplied.

Columbia linguist John McWhorter made the sharper version of a related point this week. His argument was not that Arday was a con man. He doubted that. His argument was comparative. Present a hiring committee with a thin publication record and a life story like Arday’s: a brain tumor and stroke survived the day of a dissertation defense, memory of the dissertation itself erased, armed masked intruders unrecorded by any camera, marathon times posted on an injured leg at a pace few uninjured elite athletes reach. Ask whether a candidate presenting that same file, absent the specific biography that made him a symbol, gets a Cambridge chair without serious pre-vetting first. What did the elevating, in McWhorter’s account, was the audience’s need to be seen elevating it. That need is a Girardian mechanism too. It just runs in the opposite direction from the one now operating on Cofnas: a community converging around a figure to feel good about itself, rather than converging against one to purge its tension.

Both directions share the same defect. Neither one is actually about the facts.

This is not a story about one plagiarism case, or even about how universities vet candidates. It’s a story about what institutions optimize for when the truth becomes expensive to hold. Cambridge and Ghent’s own processes, run by credentialed people with every professional incentive to get it right, converged twice on the version of events that protected social equilibrium over the version supported by the evidence. That is not a corruption problem. It’s a knowledge problem, the same one Hayek described in markets: no central process, however well-intentioned, reliably outperforms the distributed, adversarial, uncomfortable business of letting inconvenient facts surface and stand. An institution that suspends the fact-finder while leaving the finding unrebutted has not protected truth. It has told everyone watching what a career actually costs.

The next person who finds an inconvenient fact and has to decide whether to publish it now has real data on what happens next. So does everyone watching who hasn’t found one yet, and won’t go looking.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.

It can get smaller than a county. Near Maysville, Kentucky, developers offered Delsia Bare and her mother $26 million for farmland needed to anchor a 2.2-gigawatt hyperscale project. They turned it down. Neighbors down the road who’d have taken a payout from the completed project are furious that two women’s refusal now means nobody gets paid. A hyperscale data center needs a contiguous tract, and one landowner who says no can veto the payday for everyone standing next to her. Jurisdiction, it turns out, goes all the way down to a single fence line.

Before Musk left the ground, he had already found the machinery’s limit. His Colossus data center in Memphis runs on power from dozens of gas turbines sited a few hundred feet away, on the other side of the Tennessee-Mississippi line. Memphis residents fought the turbines for a year. So the newest ones went up just across the state line in Southaven, where Mississippi’s permit board approved all 41 unanimously, three weeks after the public hearing. The computers themselves stayed put in Memphis, where power is among the cheapest in the country. Then Mississippi handed xAI the largest economic development deal in state history to build the next building: $20 billion, with a permanent exemption from sales and use tax on the equipment inside it. Turbines where the permit comes easy. Servers where the power is cheap. Chips where the tax is zero. Every piece of the project sits in whichever jurisdiction charges the least for it. Neither state got the whole thing. Both gave him their best offer anyway.

That still requires a state line to exist, and someone to notice which side of it is cheaper. Orbit doesn’t require even that.

At a JPMorgan investor event on June 5, Musk was asked about putting data centers in space. “In fact, we think it’s quite easier than our communication satellites,” he said. “The Starlink V3 communication satellite is an incredibly complex machine. The AI data center would be much simpler by comparison, because it’s really just solar power and a radiator.” In orbit, that solar power runs uninterrupted, no night, no clouds, no grid to wait on. Three days later, SpaceX revealed why the rest of the claim wasn’t bravado either. AI1, the company’s first orbital compute satellite, is built on the Starlink V3 “bus,” the standard chassis and power platform SpaceX already mass-produces. The hardest problem in space-based computing, dissipating heat with no air to carry it away, gets solved by a deployable liquid radiator, drawing on hardware SpaceX had already built for an unrelated product. What was left over from one program became the whole design of the next.

No county commissioner’s district ends at the property line of a spacecraft. No state permit board can hold a hearing on an orbit. No landowner can hold out for a better number, because there’s no offer coming and no fence line to defend. The jurisdiction-shopping that worked in Memphis and Southaven still needed a jurisdiction on both ends. This needs none.

It cuts the other way too. Loudoun’s fiber-era gamble now covers 95 percent of the county’s operating budget, the difference between an 80-cent and a dollar-fifty property tax rate for every homeowner in it. None of that reaches a satellite. There is no assessed value on a spacecraft, no operating budget it can be made to fund, no $26 million falling from the sky onto anyone’s land. A county that blocks a terrestrial data center forecloses the one thing Loudoun proved was on the table, and gets nothing in exchange, because the alternative it pushed the demand toward was never going to pay anyone’s property tax, or anyone’s neighbor, to begin with.

Here is the part worth sitting with. None of this, Memphis or orbit, was planned as an end-run. Starship exists to cut launch costs, for reasons that predate any data center. The Southaven turbines exist because Tennessee has cheap power and Mississippi has fast permits, not because Musk drew a map of state lines in advance and worked backward. Musk is one step ahead of the planners, but not because he outplanned them.

Economist Israel Kirzner spent a career on a simple idea: markets move forward not because someone calculates the optimal allocation of resources, but because someone notices an opportunity sitting in plain sight that everyone else walked past. Alertness, he called it, not effort and not genius. It is not the same as having a plan. Musk did not build Starship to eventually house a data center in orbit, any more than he built a Memphis supercomputer to eventually discover that Mississippi’s permit board meets faster than Tennessee’s. He built a cheaper rocket. He needed power fast in Memphis. The opportunities showed up afterward, visible only to whoever was already holding the capability that revealed them. Hayek’s older point about knowledge sits underneath this: the relevant facts do not exist anywhere in advance, waiting for a planner diligent enough to collect them. Kirzner’s addition is what happens next. Someone has to be standing close enough to notice when they appear, and act before anyone else does.

Regulation cannot anticipate a discovery that has not happened yet. That is not a loophole. It is the ordinary condition of a world where knowledge is made, not issued.

The activists most committed to stopping data centers are, without intending it, selecting for the version of AI buildout least reachable by anyone, including themselves.

Enduring economic lessons plucked from the headlines, seen through an Austrian lens. Ground truth from 40 years in the Colón Free Zone.