Hello, dear readers!

The U.S. government owns 10% of Intel. The President is personally calling its customers, and the stock has more than quadrupled. By the only scoreboard Washington keeps, the pet project is working. So everyone is asking whether Intel stays fixed, but that's the easy question. Below, I dig into the harder one: what else got built while Washington was busy rebuilding a chipmaker?

I also look at the most comforting story in economics right now: that AI will quietly backfill the workers America is about to lose to retirement. A Yale economist compared the jobs AI can actually do with the jobs retirees are leaving, and they don't match. Here's the thing: that mismatch is about to land in somebody's workforce plan. Possibly yours.

Plus: Anthropic asks 9,700 workers how much of their job AI can already do, Google puts a Gemini number on brand advertising, a randomized trial finds AI-assisted teaching can quietly backfire, and a crew of twenty-somethings in Lockhart, Texas, is trying to build an American Shenzhen out of shipping containers and barbecue.

This issue features the usual bunch of AI Insights and recommendations for the week's podcasts, books, shows, charts, and tweets, followed by a final chuckle.

Let’s get going.

Industry Highlights: Washington Fixed Intel. Look What Else It Built.

Tim Cook went to Washington last summer to stop a tariff.

The administration threatened to impose 100% duties on every imported semiconductor, which would have hit Apple exactly where it lives. Cook worked the meetings, pledged hundreds of billions more in U.S. investment, and won his exemption.

Somewhere in those same conversations, President Trump and Commerce Secretary Howard Lutnick raised a company Cook had not come to discuss: Intel. Use their fabs, they suggested, the chip factories Intel still owns on American soil. Make some of your chips there.

The Wall Street Journal's Robbie Whelan connected those two threads this month, and nobody had reported the link before. Read it once, and it looks like a lobbying story; read it again, and you realize it's a sales call placed by the President of the United States, on behalf of a company he owns ten percent of.

Call it state capitalism, American-style. Washington now sits on Intel's cap table, in its sales meetings, and on the tariff schedule that decides what its customers pay. And the uncomfortable part isn't the arm-twisting; it's that the arm-twisting worked.

What $9 Billion Bought

The stake came first. Last August, the government converted $9 billion in federal grants into roughly 10% of Intel's equity, making Washington the largest shareholder in an American chipmaker.

It's worth remembering where that money came from. The grants were Biden-era CHIPS Act dollars, so the novelty wasn't the spending; it was turning the spending into ownership, the same debt-to-equity maneuver Republicans spent 2009 calling socialism when the patient was GM.

Then the money and the customers started arriving. SoftBank put in $2 billion in late August. Nvidia followed in September with $5 billion and an order for Intel's custom data-center chips, a partnership Jensen Huang called historic. In April, Musk folded Intel into Terafab, his plan to design, fabricate, and package ultra-high-performance chips at scale. Google Cloud placed a large order for Xeon CPUs, the general-purpose processors that do the ordinary work surrounding an AI model. And this month, Trump announced on Truth Social that Apple will use Intel-made chips, reportedly in both Macs and iPhones, sending the shares to record highs.

Intel executives are blunt about what the money did. Without the federal conversion, Nvidia's $5 billion and SoftBank's $2 billion, they would have had to gut capital spending. Instead, CEO Lip-Bu Tan held capex flat and redirected it away from pouring new fabs and toward tooling the plants Intel already had, so the company could build more of what customers were actually buying.

The Access Nobody Else Gets

The involvement runs well past deal-brokering. Tan flies to Washington roughly once a month and talks by phone with Lutnick regularly, briefing him on customer relationships and business conditions. Bill Frauenhofer, the administration's chips czar and a career semiconductor banker, takes a quarterly briefing from CFO David Zinsner, and his staff meets Intel executives in Washington and at Santa Clara to track manufacturing progress firsthand.

Can you name another 10% shareholder with that kind of access? And one who can also call your biggest customer and ask for the order?

The Man in the Middle

Here's the part the comeback narrative skips: a year ago, the president wanted Tan fired.

Senator Tom Cotton went after him publicly over his China ties, decades of investments through his venture fund, Walden International, and Chinese customers during his years running Cadence Design Systems. In early August, Trump posted that Tan should resign. Weeks later, Tan was sitting in the Oval Office explaining that he was not a Chinese spy.

He walked out having converted the president into a patron. Trump decided Tan was a winner and floated the government stake himself. From "resign" to "largest shareholder" in three weeks and one meeting.

To be fair to Tan, Washington deserves only part of the credit. At 66, he took what may be the hardest job in tech instead of retiring (his own framing on the No Priors podcast) and has pushed a famously slow company toward faster decision-making, greater customer responsiveness, and greater engineer accountability. He poached executives from Samsung and SK Hynix, hired Cadence's top silicon engineer to run a new Central Engineering Group, and pointed the roadmap at the CPU's place in agentic AI and inference, the running of models rather than the training of them.

Luck cooperated, too. This phase of the AI boom eats CPUs, which happen to be Intel's specialty, and data-center revenue rose 22% year-over-year to $5.1 billion in April.

But the culture change is showing up where it counts. Google Cloud's Mark Lohmeyer says Intel now listens to feedback and executes on it, which makes Google keener to partner. Of course, that compliment tells you plenty about the old Intel as well.

The Other Side of the Ledger

Intel is not fixed. Intel Foundry, the arm that manufactures chips to order for other companies and the half of the business Washington actually cares about, lost $10.4 billion over the last four fiscal quarters. The company posted a $3.7 billion net loss in the same quarter that data-center sales jumped 22%. Outside customers spent years learning not to trust the foundry to deliver usable wafers in volume, and that reputation does not reverse on a Truth Social post.

Both Intel and the government are betting on advanced packaging, arrays of small chiplets combined to work as a single custom semiconductor, as the best route to competing with TSMC. Zinsner told January's earnings call that the revenue there was arriving faster than even he had expected. Encouraging, though I'd point out that's a CFO's adjective, not an audited number.

The Real Risk Isn’t Technical

Now suppose the engineering all works: advanced packaging lands, the foundry stops bleeding, the customers stay. Intel is still carrying an exposure that no roadmap addresses.

The Cato Institute's Scott Lincicome put it this way: "Being the government's darling only works as long as you're doing well." Patronage isn't a contract; it's a mood, and it belongs to whoever holds the office next. Politicians run on two-year clocks while fabs run on ten-year ones, which means Tan is building on a timeline that outlasts every person currently helping him.

The Bottom Line

The scoreboard says this worked. The 10% stake has quadrupled in value, an immediate paper return for taxpayers, and Intel is more credible today than at any point in a decade. If you believe the world's most advanced fabs sitting inside Chinese missile range is a national-security problem, this was money well spent, and cheaper now than later.

But look at what actually got built here. Not just a chipmaker, but a template: a president can now attack a CEO on social media, take a slice of his company three weeks later, and personally call the world's most valuable firms to fill his order book. That machinery doesn't get dismantled when the administration changes, it gets inherited.

Intel's turnaround is real. Whether the precedent reassures or alarms you probably has less to do with the economics than with who you picture holding the phone next.

The Future of Work: Aging and AI Won’t Cancel Each Other Out

Washington can convert grants into equity. It cannot convert them into people. Somebody still has to run the fabs, and that is where this gets harder.

There’s a comforting story going around, and you have heard some version of it in a board meeting. America is aging. The birth rate is falling. Too few workers? We’ll have AI. Older workers slowing down? AI. Not enough caregivers? AI. Two crises, one solution, no action required.

Martha Gimbel, who runs the Budget Lab at Yale, took that story apart in a Bloomberg essay this month. Her argument is simple and unwelcome: two enormous transitions arriving at the same time don’t cancel out just because they point in opposite directions. In the short run, the better bet is that they compound.

The Math Isn’t Subtle

The Congressional Budget Office projects that annual U.S. deaths will exceed births starting in 2031, which leaves immigration as the only thing keeping the population growing. Fertility is projected to fall to 1.5 births per woman by 2055.

History helps less than you'd hope, because nearly every past technological upheaval happened while populations were exploding. England's population doubled between 1750 and 1814, and the economic historian Joel Mokyr has simulated a counterfactual in which the Industrial Revolution would have raised incomes immediately if not for all those extra people. Run that experiment in reverse, and you get the optimistic case: transformative technology, a tight labor market, rising wages.

There's even some evidence for it. Industrial robots appear to have blunted the productivity drag of aging in South Korea, and Daron Acemoglu and Pascual Restrepo have shown that aging creates the labor shortages that push firms to automate in the first place.

Then again, it may cut the other way. One study found that Japanese executives under 50 were 23% more likely to adopt AI than their older colleagues; an aging country may simply be slower to adopt the technology at all. Neither can be true, but notice what they share: in either version, demographics is steering AI, not rescuing us from it.

The Jobs Don’t Line Up

Now for the finding that should change how you plan. Gimbel's occupational data shows almost no correlation between the jobs with the oldest workforces and the jobs most exposed to AI. Passenger attendants on planes, trains, and ships skew old and are nearly impossible to automate, while medical transcriptionists are heavily exposed and skew young. There are exceptions, but the overall relationship is weak. AI does not politely slot itself into the roles aging is emptying out.

In theory, the transcriptionist retrained as a flight attendant. In practice, that is precisely the transition the China shock proved brutal (factory towns lost their industry after China entered the WTO and mostly never got it back), especially when the losses cluster in one region. And the workers best equipped to make that jump, the young ones, are exactly who a shrinking country has fewer of. The economy needs maximum flexibility at the moment, it has the least.

What This Means for Leaders

I saw a version of this up close just in the past week. Two CPOs I know are retiring from different industries, and a third who thought his run was over just landed another CPO seat that matches his background to a tee. It's always heartening to see companies that understand the value a seasoned executive brings, because what they're really buying is the institutional knowledge no tool can replicate: which supplier will actually pick up the phone at 2 a.m. when a line goes down, which contract clause got added after the last crisis, and why. It doesn't matter what generation you belong to; if you can add value, you'll always be in demand. The trouble is that for every company making that hire, plenty more are penciling in the retirement wave, pointing at an AI pilot, and calling it a workforce plan. 

The AI tools landing on our desks can draft an RFP and summarize a contract in seconds, and that's genuinely useful. But none of them know your suppliers. That knowledge doesn't transfer through a chatbot; it transfers through hiring and mentoring, and I'm not seeing nearly enough of either.

Here's the thing: economists mostly don't argue about the destination. Higher productivity eventually shows up as growth, and governments have tools to spread it around. They argue about the transition. Everyone reaches for the Industrial Revolution as the reassuring analogy, and living standards did rise, eventually. Living through it was disruptive, violent, and long, with population growth working in its favor. This one doesn't.

I've read enough workforce plans to tell you the aging assumption is almost never argued. It just sits in the model, load-bearing and unexamined: attrition rises, hiring need stays flat, because AI. Nobody writes it down as a bet, but that's exactly what it is.

So find where that assumption lives in your own plan, and make somebody defend it out loud. The tasks AI absorbs and the roles your retirees vacate are unlikely to be the same tasks, and the gap between them isn't a technology problem you can procure your way out of. It's a hiring, training, and geography problem, and it lands on your desk, not on a model.

AI Insights

Anthropic’s Economic Index: June's report linked a survey of roughly 9,700 Claude users to their actual usage data, and more than 35% expect AI to handle most or nearly all of their job tasks within twelve months. The counterintuitive part is who's optimistic: the heaviest delegators report the best expectations for their pay and job security, not the worst. Of course, read the sample before the headline. Everyone surveyed is already an active user, and the workers' AI is replacing rather than assisting, so it doesn't generate sessions to sample. 

Google: A new Ads metric called Qualified Future Conversion uses Gemini to forecast the revenue expected within 180 days of an ad interaction, tying early signals like brand searches to eventual purchases. Brand advertising has resisted measurement for a century, and now it's finally being measured by the company selling the ads. What could possibly go wrong?  

AI in the classroom: One of the first randomized trials of AI in real classrooms (193 teachers and 2,800 students at a school chain in Turkey, run by Wharton's Alp Sungu with Angela Duckworth) handed half the teachers a ChatGPT assistant tuned to the national curriculum. Students in those classes rated the material less interesting and less important, and where the teacher was already weaker, achievement and confidence both fell on externally graded exams. Average scores didn't move, so this isn't a simple "AI ruins school" story. The sharper finding is that strong teachers treated the output as a first draft while weaker ones shipped it as-is. Still a working paper, but worth sitting with.

The Supply Aside

📕 Read: How to Get Rich in American History

Moore's thesis is this whole issue in miniature: wealth is built, not found, and the portal to the American dream is not a stock-trading app. He likes debt, pointing out that a young Ben Franklin borrowed heavily to start his printing business. He thinks inherited wealth is overrated and rarely lasts. He shrugs at diversification until you're already rich. And he wants you doing something with your hands: take in boarders, start a business, buy a property, and improve it. The real money, he writes, is active.

Roger Lowenstein, who wrote the Buffett biography, reviewed it for the WSJ and did not hand out a rave. He finds the history glib, catches errors (a greenback dated a year late, a fumbled Berkshire fact), and notes the book wants it both ways: distrust the experts, but trust the best gurus. He still credits its originality and its human streak. Moore's own stated regret is how much attention he paid to Jim Cramer while his daughter was learning to walk.

Read it for the argument, and keep a fact-checker beside you for the history.

What Else I’m Reading

The secret to good questions (The Economist, Bartleby): By the time you reach a boardroom, your principal output is questions, which is a decent reason to get better at them. Georgetown's Karen Huang finds that people who ask more questions are better liked. Researchers at ETH Zurich, trawling Glassdoor, found that candidates are likelier to turn down high-paying offers after an easy interview; softballs read as a signal about your future colleagues. And Columbia and Berkeley researchers trained a model to spot "curveball" questions on earnings calls, the kind that are hard to predict and too relevant to dodge. Those moved share prices, and the people asking them were either star analysts or analysts brand new to the company. Apparently, deep expertise and fresh ignorance both beat the middle.

Why recruiters can’t find workers and new grads can’t find jobs — it’s not AI (The Washington Post / The Hechinger Report): The reported twin of Gimbel's essay. Between 2024 and 2032, more than 18 million college-educated workers leave the labor force while fewer than 14 million enter it, a gap Georgetown puts at 4.6 million and Lightcast puts at 6 million. It lands on nurses, teachers, engineers, airplane mechanics, and construction workers: the jobs AI can't do. And the tie back to our lead story is uncomfortably direct. Semiconductors are projected to add nearly 115,000 jobs by 2030, roughly 67,000 more than the technicians and engineers who exist to fill them. An equity stake takes an afternoon. An engineer takes eighteen years.

Chinese automakers are building a quiet presence in the U.S. (Jalopnik): Our reindustrialization theme, running in reverse. Chinese carmakers can't sell here (the tariff is 102.5%), yet Automotive News found them parked on American soil anyway. SAIC keeps an office eight miles north of Detroit behind a darkened, locked glass door: logo on the glass, nobody in the phone directory. Nio's San Jose research center answers with a machine that reindustrialization theme, running in reverse. Chinese carmakers can’t sell here — the tariff is 102.5% — yet Automotive News found them parked on American soil anyway. SAIC keeps an office eight miles north of Detroit behind a darkened, locked glass door: logo on the glass, nobody in the phone directory. Nio’s San Jose research center answers with a machine that won’t say the company’s name. AlixPartners’ Stephen Dyer says they’re here to track the industry, hunt talent, and learn the supply base. Canada just agreed to admit up to 49,000 Chinese vehicles at a reduced tariff, and U.S. rules broadly align with Canada’s. Washington is pushing Intel to bring chipmaking home while Chinese automakers wait out the wall from an office park outside Detroit.

📺 Watch: How a Texas Startup Ranch Plans to Save American Manufacturing

Ashlee Vance's Core Memory takes the first outsider's look inside Proto-Town, a 1,200-acre ranch outside Lockhart, Texas that a group of twenty-somethings is trying to turn into an American Shenzhen. Residents live in Conex trailers and one ranch house, work morning to night, and build the things software people won't touch: solar-powered air conditioning, a self-cleaning desalination centrifuge, autonomous excavators from Bedrock Robotics, drones the size of a basketball court. A nuclear startup is standing up a research reactor for medical isotopes, and a semiconductor startup has moved in. Twelve companies so far, on a $20 million seed round from Bill Ackman, Josh Kushner's Thrive Capital, and Coinbase co-founder Fred Ehrsam that valued the place around $100 million.

One detail made me sit up. In 2022, Lockhart was a finalist for a $100 billion Micron fab, and Micron chose upstate New York instead. The town that lost the chip plant is now hosting the people trying to rebuild American hardware without asking Washington for anything.

That's the pairing worth your twenty-two minutes. Intel is top-down: federal equity and a president working the phones. Proto-Town is bottom-up: private money, cheap land, loose rules, and a bet that a Y Combinator for atoms beats industrial policy. Of course, both are frightened of exactly the same thing, which is that the fabs, the tooling, and the supply base all sit on the wrong side of the Pacific. Come for the hardware, stay for the barbecue.

👂 Listen: All-In: OpenAI CFO Sarah Friar

The Intel story from the demand side. Washington is spending public money to rebuild the capacity to make the chips; Friar is the one writing $100 billion-plus in checks to rent what those chips become. Chamath, Jason, Sacks, and Friedberg push her on OpenAI's IPO timeline, the arms race with Anthropic and Google, the compute crunch, and a new hardware device. The bottleneck she describes and the bottleneck Washington is trying to fix are the same bottleneck, just seen from opposite ends. A tight thirty-two minutes, and well worth the listen.

🧠 Think: Whose Footsteps?

Everyone building something ambitious right now names it after something that already exists: the next Shenzhen, the next Silicon Valley, the next Detroit.

I understand the impulse. A comparison is a shortcut through a conversation you don't have time to have. But here's the thing: Shenzhen isn't a place with cheap land and good vibes. It's four decades of accumulated suppliers, a machine shop that will run you 400 units by Thursday, and thousands of people who already failed at this once and stayed anyway. You can't buy that; you can only put in the time.

A company is no different. Intel's problem was never a shortage of money or attention. It was a decade of decisions, and no shareholder, however powerful, gets to skip the rebuilding on anyone's behalf.

Basho is this issue's quote for a reason: don't seek to follow the footsteps of the wise, seek what they sought.

Charts of the Week

The Supply Times Analytics: Chargebacks have climbed every year since 2021 and are closing in on 400 million globally, with the U.S. accounting for roughly 40% of the volume. Some of that is honest confusion at the level of the statement. A meaningful share is “friendly fraud,” which is customers disputing purchases they made and received. Either way, read it as a fast-growing tax on e-commerce, and one that lands on the merchant rather than the bank.

The Supply Times Analytics: SpaceX sits around $2 trillion, while sell-side price targets run from Goldman’s $2.7 trillion to Raymond James’ $10.5 trillion. Look at what fits inside that spread: the entire listed stock markets of Germany, France, and the U.K. When analysts covering one private company disagree by more than a G7 equity market, the number being published is a mood, not a valuation.

The Supply Times Analytics: The McKinsey Health Institute surveyed 30,000 employees across 30 countries and found that those reporting a traumatic life event scored lower on all six performance measures: engagement down 10 points, adaptability 7, and psychological safety 6. The relationship is descriptive, not causal. But adaptability is the one to watch, given everything above. The workforce we’re asking to absorb the AI transition is not a fresh set of batteries.

Quote of the Week

❝

“Do not seek to follow in the footsteps of the wise; seek what they sought.”

— Matsuo Basho

Tweet of the Week

The Final Chuckle

Thanks so much for reading. I’d love to know what you think about this issue and how I can make it more useful to you. If you have suggestions or topics you want to see me address, email me at [email protected]!