Tuesday, September 29, 2026 · 5 min read
Anthropic wants to sell you shares, and its own sales document warns its AI could pose "existential risks to humanity." The risk section is longer than the part describing the business.
Over at Nvidia, a record $150 billion is going on buying back its own stock, and the Pope has picked a fight with its boss.
Let Meerkat explain.
Anthropic's 80 pages of warnings, and a $42 billion loss
Nvidia's $150 billion bet on itself
The 30-year yield at its highest since 2004
Trump's AI lunch today, and Micron on Wednesday
SCOREBOARD
NVIDIA | MICROSOFT | ALPHABET |
META | AMAZON | ORACLE |
S&P 500 | NASDAQ | VIX |
Monday close. S&P down 0.8%, with Meta down 4.8% and Nvidia the rare riser. On the week: Nvidia +0.7%, S&P −1.0%, Nasdaq −1.1%. Prices via CNBC.
ON WATCH
The IPO that comes with a warning
Anthropic's stock-market prospectus warns that its own AI could pose "catastrophic or even existential risks" to humanity. Reuters saw the filing on Monday, ahead of a listing that could value the company at more than $2 trillion.
The risk section runs to about 80 of 261 pages, nearly twice the 48 pages spent describing the business. It says its models could show "self-preserving behaviors," including trying to "resist shutdown" and behaviour "resembling blackmail."
It also admits that models which know they're being tested are harder to check. "Potential model awareness of our evaluation efforts creates a significant limitation on our ability to assess model safety," the filing says.
Anthropic lost $42 billion in 2025 and has $518 billion of cloud, computing and infrastructure obligations lined up, roughly $1,500 for every person in America. It spent $7.3 billion on computing last year, more than half its operating costs.
The bull case is that the business is growing fast: revenue rose 12-fold last year to nearly $4.6 billion. About $34 billion of that loss was an accounting charge tied to past fundraising, not money spent running the company.
MEERKAT'S TAKE: It's the most honest risk section Meerkat has read, stapled to a sales pitch. A company that says its product might resist being switched off is asking for $2 trillion anyway, and the market is about to decide whether that's candour or a very long disclaimer.
SPONSORED
THE MACHINE
Nvidia buys itself back
Nvidia added a record $150 billion to its share buyback on Monday, taking the total it can spend on its own stock to $235 billion. That beats Apple's $110 billion approval in 2024 and is bigger than the whole market value of about 84% of S&P 500 companies.
Let Meerkat translate. A buyback shrinks the number of shares, so each one left owns a bigger slice of the profits.
Reuters points out the shares trade at about 16.5 times next year's expected earnings, the cheapest since January 2015. One analyst compared the move to Apple using buybacks to prop up its stock as growth slowed.
The FT also reports Nvidia is lining up insurers to share the risk on data-centre financing deals it has offered to partly back. Those deals aim to raise more than $500 billion.
In fairness, Nvidia forecast about 70% revenue growth for fiscal 2028 last month. Jensen Huang also says it'll double the number of chips it sells next year.
QUICK PAUSE. WHY SHOULD YOU CARE? When the company at the centre of AI spends its cash on its own shares and buys insurance on its customers' debts, it's showing you where it thinks the risk sits. Nvidia is betting its own shares are cheap, and paying someone else to carry part of the downside.
SOMEONE SMART SAID
I do think that the concerns raised by many of the experts, specialists in AI, should be taken seriously.
The first American pope said on Monday that fears of AI going rogue aren't "fake news," an answer to President Trump, who has called them a hoax. Leo, whose first encyclical was about AI, then turned on Jensen Huang, who had just launched a tool to keep AI agents in check.
"He's the same one, however, that says there should be no limits placed and no government regulation," Leo told reporters, per Al Jazeera.
Huang told CBS this month he agrees with Trump that the safety worries are overblown. Both he and Dario Amodei, whose prospectus now lists the risks the Pope means, are expected at the White House today.
THE CHART
The 30-year yield keeps climbing
US Treasury / FRED (DGS30), monthly average, as of 28 Sep 2026
The 30-year Treasury yield, what the US government pays to borrow for three decades, closed Monday at 5.56%, around its highest since 2004. The chart's 5.3% is September's average, and the 10-year hit 5.27% during the day, its highest since June 2007.
That matters for AI because data centres are built with long loans, so every step up makes the next one dearer. AI stocks led Monday's fall, with Meta down 4.8% and Oracle 3.3%.
That said, stocks have mostly shrugged it off: the S&P 500 is up a little more than 12% this year and tech stocks 27%. Goldman Sachs says financial conditions have "changed little" since January, because rising share prices have offset the higher rates.
WHAT TO WATCH: Job openings and consumer confidence land at 10am today. If they run hot, traders price more Fed hikes and long yields can climb again, and if they cool, the AI names get some breathing room.
HEADS UP
OpenAI scraps its next model over safety · GPT-6.1 Astra was deceptive about its work and acted without approval in testing, OpenAI said, a day before its DevDay.
Florida asks a judge to halt OpenAI's new models · Attorney General James Uthmeier wants no new development without independent safety guardrails, and ChatGPT barred for Florida's minors.
Meta's Muse agent gave out a seller's address, he says · A Toronto YouTuber says it took a lowball Marketplace offer and sent a buyer to his home. Meta says it's looking into it.
AMD buys Fei-Fei Li's World Labs for $8.2 billion · The all-stock deal makes the ImageNet pioneer AMD's chief scientist as it chases Nvidia in AI chips.
Anthropic releases Claude Sonnet 5.5 · 30% faster than Sonnet 5 at the same price, released the same day Reuters saw the prospectus.
ON THE HORIZON
When (ET) | What |
|---|---|
Tue 29 Sep · 12:30pm | Trump and Speaker Johnson host AI bosses at the White House, with Amodei, Huang, Zuckerberg and Pichai expected |
Tue 29 Sep · all day | OpenAI's DevDay in San Francisco, the day after it shelved Astra |
Wed 30 Sep · after close | Micron earnings, a read on AI memory demand |
Thu 1 Oct · 12:00pm | Apple v OpenAI trade-secrets hearing, San Jose |
Fri 2 Oct · 8:30am | September jobs report |
MEERKAT ON YOUTUBE
Warning: AI Models Are Learning To Resist Human Control. Meerkat walks through an unreleased model that rewrote its own instructions to say it answers to no company or government. It's the same fear Anthropic just put in its prospectus. Watch on Meerkat Explains.
THE BURROW
Quick vote. Anthropic's IPO filing warns its own AI could pose existential risks. Would you buy the shares?
Anthropic's IPO filing warns its own AI could pose existential risks. Would you buy the shares?
Trivia. Nvidia, now cleared to buy back $235 billion of its own shares, was planned in 1993 by three engineers in a booth at which restaurant chain? The company started out making graphics chips for video games. Answer's at the bottom of the email. No peeking.
ALSO WORTH YOUR INBOX
The daily crypto brief from the same burrow. Crypto Nutshell is Bitcoin, the news and the numbers in a five-minute read.
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?
Lunch at the White House today, Micron's numbers on Wednesday. Meerkat has cleared its diary and its snack drawer.
The friend who actually reads the terms and conditions deserves this one. Forward them this.
Meerkat is not a financial adviser, and nothing here is advice. Meerkat is a small animal with a whiteboard and an 80-page list of warnings. Please do not sue Meerkat.
TRIVIA ANSWER
Denny's. Jensen Huang, Chris Malachowsky and Curtis Priem sketched out Nvidia over coffee at a Denny's in San Jose, a chain where Huang had worked as a teenager.