Monday, September 21, 2026 · 5 min read
Together with
Four people who pay for ChatGPT and Claude just did something odd: they sued the companies for agreeing to slow AI down. The complaint says the industry's big "safety" pledge was really an illegal deal to make the product worse.
Meanwhile Nvidia's boss says the doom driving that pledge is fiction, and Meta is quietly building its own chips so it needs Nvidia less.
Let Meerkat explain.
Why four subscribers are suing the AI giants for agreeing to slow down
Meta's own AI chip, and the bet against Nvidia underneath it
The AI buildout's growing pile of debt, in one chart
The week ahead, from a Xi state dinner to Micron's chips
SCOREBOARD
NVIDIA | MICROSOFT | ALPHABET |
META | AMAZON | ORACLE |
S&P 500 | NASDAQ | VIX |
Friday close. S&P up 0.2%, a quiet finish to a flat week. On the week: Nvidia +1.82%, S&P −0.08%, Nasdaq +0.72%.
ON WATCH
The slowdown goes to court
On Thursday four people who pay for ChatGPT, Claude, Grok and Gemini sued the companies behind them for agreeing to make the product worse. The complaint, Buist v. Anthropic, filed 18 September in San Francisco, says OpenAI, Anthropic, Google and Elon Musk's xAI illegally teamed up to hold AI back.
The evidence is public. On 12 September Dario Amodei published an essay, "We Must Pace the Frontier," urging rivals to slow down for a year or two so safety could catch up, and within the hour Altman, Musk and Google's Demis Hassabis all agreed.
Antitrust law treats rivals agreeing to hold back a product as close to automatically illegal, and the suit wants triple damages for everyone who pays for a chatbot.
The case against reading too much into it: Amodei's essay admitted the coordination would need an antitrust waiver, so the labs saw it coming and are betting a safety pact isn't a price-fixing ring. No judge has ruled, and Anthropic has offered to open its work to outside inspectors.
The pledge grew from a fear this newsletter opened with: former Anthropic researcher Jacob Coxon warning that AI's own builders think it could kill everyone this decade. Now a court decides whether slowing for safety is a public service or a cartel.
MEERKAT'S TAKE: There's something very 2026 about suing an AI company for promising to be careful. Whether or not this case wins, it hands every lab a reason to stop cooperating on safety, which is a strange thing to be cheering for.
SPONSORED
Your Next Side Hustle is Already on This List
Ready to stop scrolling job boards and start building something of your own?
The Hustle's free "Side Hustle Ideas Database" uncovers 100 unconventional, actually-actionable ways to earn on the side.
Inside you'll find:
A curated list of 100 real side hustle ideas spanning e-commerce, services, content, and weird corners of the internet most people never think to look
Opportunities at every level, whether you have 5 hours a week or 50, zero capital or a small runway
Offbeat, under-the-radar ideas you won't find in generic "how to make money" listicles
Subscribe for free today and get instant access. Your next income stream is one click away.
THE MACHINE
Meta builds its own chip
This month Meta started mass-producing its first serious in-house AI chip, codenamed Iris, the payoff of a silicon program it's run quietly for years. The point is blunt: stop handing quite so much money to Nvidia.
Meta expects to spend as much as $145 billion on AI this year, much of it on Nvidia's famously high-margin chips. Iris, designed with Broadcom and built by TSMC, is meant to run Meta's own ads and feeds more cheaply, though it'll keep buying Nvidia GPUs by the truckload too.
Meta isn't alone. Google, Amazon and now OpenAI are all rolling their own silicon, and every chip they make in-house is one they don't buy from Nvidia, the company the entire AI trade is leaning on.
To be fair, this is a slow game and Nvidia is still winning it easily: custom chips handle narrow jobs, its GPUs still do the heavy training, and demand is so high the giants are buying both. The threat is to Nvidia's margins over years, not its business tomorrow.
QUICK PAUSE. WHY SHOULD YOU CARE? Nvidia became the most valuable company on earth on the belief that everyone has to buy its chips. The day its biggest customers start making their own in volume is the day that belief gets tested.
SOMEONE SMART SAID
2030 is not going to be the end of the world. There is 0% chance that's going to be the end of the world.
Huang said this to CBS on Friday, swatting away a run of warnings from inside the industry that AI could wipe out humanity by the end of the decade. He called the doom "not grounded in science" and said scaring people is "irresponsible," and he wants no new AI laws either.
Remember who's talking, though. Huang sells the chips the whole boom runs on, so a calm, keep-buying message suits him nicely.
But he's answering a real argument, not a straw man. The same fear he's brushing off is the one that pushed Amodei, Altman and Musk to pledge a slowdown, the pledge now being fought over in court. Whether the risk is 0% or something worth a pause is, weirdly, the question the whole industry is arguing and litigating at once.
THE CHART
The buildout runs on debt
Outstanding data-center securitizations, $B. Source: SFA Research and Barclays, 2026.
The physical side of AI, the warehouses stuffed with chips, increasingly runs on borrowed money. Bonds backed by data centres have jumped from around $4 billion a few years ago to roughly $61 billion, as the spending races ahead of even the hyperscalers' enormous profits.
The counter is that most of the biggest spenders can still build from cash, and lenders are keen because the tenants are the richest companies on earth. The worry is the newer, debt-heavy players, and how fast the pile is growing.
WHAT TO WATCH: If the AI names keep delivering, this debt stays boring and gets repaid. If the returns disappoint, the money that funded the warehouses is the first thing to vanish, and a half-finished data centre is a very expensive shed.
HEADS UP
Researchers used Anthropic's Claude to break into OpenAI in under 72 hours · They chained two bugs to reach OpenAI's internal code, left a "harmless" pull request as proof, and pocketed a $6,500 bounty.
Tech titans head to Trump's state dinner for Xi · Nadella, Cook, Huang, Altman, Bezos and Musk are due on 24 September, with AI and the US-China chip rivalry on the menu.
Anthropic says a state group turned Claude into a cyber weapon · Its threat report describes a Chinese group using Claude Code to run a hacking campaign against about 30 firms and agencies.
AI-tied job cuts keep piling up · Outplacement firm Challenger counts more than 36,000 job cuts tied to AI in two months, even as AI-skilled hiring keeps climbing.
Oil slips as crude flows stay "surprisingly strong" · Prices fell Monday even with Middle East tensions high, as traders wait for Saudi shipments to recover.
ON THE HORIZON
When (ET) | What |
|---|---|
Wed 24 Sep | White House state dinner for Xi; Nvidia, OpenAI, Apple and Microsoft chiefs attending, AI on the agenda. |
Tue 29 Sep · 10:00am | JOLTS job openings (Aug), a read on hiring. |
Wed 30 Sep · after close | Micron earnings, an AI-memory read. |
Late Sep | Anthropic's S-1 and roughly $2tn IPO window (reported, unconfirmed). |
Wed 28 Oct | Fed's next rate decision, after this month's first hike in three years. |
MEERKAT ON YOUTUBE
TRUMP PANICS On AI: Fears MASSIVE AI Crash! Meerkat unpacks why even the White House has started worrying out loud about an AI bust, and what a burst bubble would actually do on the way down. Watch on Meerkat Explains.
THE BURROW
Quick vote. Four subscribers are suing OpenAI, Anthropic, Google and xAI for agreeing to slow AI down. Is that a real antitrust case?
Four subscribers are suing OpenAI, Anthropic, Google and xAI for agreeing to slow AI down. Is that a real antitrust case?
Trivia. Which oil giant did the US Supreme Court order broken up in 1911, in the most famous antitrust case in American history? Answer's at the bottom of the email. No peeking.
ALSO WORTH YOUR INBOX
Crypto Nutshell. The daily crypto brief from the same burrow, Bitcoin and the numbers in five minutes. Join 95,000+ readers.
HOW DID WE DO?
One tap, and Meerkat listens. How did today's issue land? Hit a button, and add a line if you want to tell Meerkat why.
How did Meerkat do today?
Monday's done. Meerkat will be back at 6:30 tomorrow, once it's finished the 40-page complaint.
Got a friend who still thinks the AI giants are locked in a race to the death? Forward them this.
Meerkat is not telling you to buy anything, sell anything, or short anything. Meerkat is not a financial adviser. Meerkat is a small animal with a whiteboard and a lawsuit it can't afford to file. Please do not sue Meerkat.
TRIVIA ANSWER
Standard Oil. In 1911 the Supreme Court broke John D. Rockefeller's Standard Oil into 34 pieces under the Sherman Act, the very law these four AI subscribers are now suing under.