Tariffs make physical inputs costlier pushing firms toward AI to cut labor. But they also raise AI hardware costs and stall data center builds. Are we running two clashing policies forcing AI adoption while kneecapping the infrastructure it needs?

A strategy at war with itself.

1. They're deploying AI to create sludge on product pages for short-term sales.

2. This sludge kills the user experience, threatening the profitability of their ad business.

3. But that ad business is the key funding source for their massive AI arms race.

Nvidia’s real power isn’t just market share, it’s time lock-in. Every major AI model is so optimized for CUDA that even a better chip can’t win without years of retooling. It’s the same play Intel ran with x86: the software ecosystem traps competitors before they start.

LEI data shows the economy isn’t matching AI-driven fund manager euphoria. Maybe AI is just another PR cycle wrapped in a bubble.

The real question: what triggers the unwind? A big earnings miss? A hot CPI print? With extreme crowding, the exit door is tiny when sentiment flips.

How can it be "humanist" if access is determined by who can pay the most?

Buy the underdog. In every key market, the #1 paper is often complacent, debt-ridden, and slow. The real opportunity lies with the #2 paper—the one with brand recognition, a chip on its shoulder, and a history of fighting from behind.

Generative AI isn’t software, it’s the first digital heavy industry—high variable costs, massive depreciating capex, and a railroad/utility path to low margins, regulation, and multiple compression.

Anthropic’s short-term operational fix is the Lehman Moment for trust in proprietary AI SaaS. They may have saved a week of GPU burn, but they just set the stage for an industry-wide OSS acceleration curve—and that curve will change the economics of AI forever.

This morning I shared some thoughts on Torsten’s commentary.

The IT bubble was a bet on Permissionless Creation.

This "AI Bubble" is a bet on Permissioned Consolidation by 5-7 monopolies.

And with it, the death of market dynamism.

Copilot saw your Broadcom tab open and made a portfolio call.

Forgive me for not taking cognitive advice from the man whose last genius innovation was an empty cartoon room.

Those wild revenue swings show the income isn't recurring—it's a string of one-off "jackpots." A jackpot business can't pay fixed GPU bills. That wasn't a strategic exit. It was an emergency evacuation from a business broken on both cost and revenue.

Meta's core business prints "boring" money, which Wall Street punishes. AGI is the ultimate narrative to launder those ad profits into a growth story, solving a valuation problem, not a tech one.

If emotions are just code, then why hasn’t the guy trying to colonize Mars figured out how to program basic humility?

Excellent piece. The precedent isn't Moviepass, it's the Broadcom/VMware playbook. A flawed model (AI's compute burn vs. Broadcom's acquisition debt) leads to the same outcome: squeezing your most locked-in customers for cash with 10x price hikes and forced subscriptions.

The $500m is the noise. The signal is the $80B GPU bill + the new Sec 174 tax hit. The layoffs are the payment.

Part 7…

This crisis isn’t an ending; it’s a filter. Thinly capitalized demo-apps will wash out. Survivors will be those that saw the foundation model as the most volatile counterparty on their cap table.