Good morning

Anthropic wants public investors to pay about 31 times revenue while its own alignment lead posts his extinction odds. Valor paid its investors in SpaceX shares instead of cash, and two Automattic executives signed each other's severance while their CEO was sidelined.

Let's get into it.

That is the valuation Anthropic is reportedly heading toward in its IPO, per Fortune's Term Sheet this morning. Rhodium Group, in estimates published the same day, pegs Anthropic's annual recurring revenue at $65 billion and its current valuation-to-revenue ratio at 21x.

Do the division. Twenty one times $65 billion is about $1.37 trillion. Asking for $2 trillion means asking public investors to pay roughly 31 times revenue, about 50 percent more per dollar of ARR than whoever bought the last private round paid. Rhodium ran the same arithmetic on Moonshot and DeepSeek, got 50x and 163x, and called those ratios exorbitant. Anthropic at 31x is the cheap one in that comparison, which tells you more about the comparison than about Anthropic.

The strange part is who is objecting. The loudest people arguing the offering should slow down work there. Evan Hubinger, who runs Anthropic's Alignment Science team, wrote that he puts the odds of AI killing all humans above 10 percent within the decade and that the company does not have a plan for aligning superintelligence. Another researcher, Drake Thomas, wrote that he would burn his equity to the ground for a 1 percent better chance of survival. At a $2 trillion listing, a tenth of a percent of the company is $2 billion, so that is a fairly expensive sentence to type in public.

You can steelman the disclosure. A company that has said since 2021 that its product might be catastrophic is being consistent when its employees say so three weeks before an S-1, and Dario Amodei has put his own p(doom) between 10 and 25 percent. Consistency is a real defense. It is also a strange one to hand a bank building a roadshow deck.

SOC Investment Group, a labor-affiliated shareholder group, has asked Anthropic to delay the offering. It will not work. The founder mechanic worth watching is the one every employee at every AI lab now faces: your equity vests on the assumption the thing you are worried about does not happen. Nobody has priced that risk yet, and the IPO is the first time anyone has to try.

D-Robotics ($400M Raised, Series C). Chips, an operating system and developer tools for robots, whose Sunrise S600 chip picked up more than 20 embodied AI customers within six months of launch. The release names its lead only as "a leading global internet company." Backed by Mirae Asset and Meituan.

Circle Pharma ($92.5M Raised, Series E). Oral macrocycle cancer drugs, with a cyclin D1 inhibitor headed for the clinic in early 2027. Backed by The Column Group and Eli Lilly.

Ravel ($8.2M Raised, Seed). Recycling for blended textiles, the polyester and cotton mixes most recyclers turn away.

Noise ($5.5M Raised, Seed). Pays everyday people to make content for brands, with 1.5 million creators signed up a year after founding. Backed by Capital Midwest, M25 and Grishin Robotics.

Aristotle ($5M Raised, Seed). Voice-first AI tutoring for teenagers, with angels from Anthropic, OpenAI and Ramp. Backed by True Ventures.

Antfly ($2M Raised, Pre-seed). Portland database that folds the five or six stores an AI retrieval stack usually needs into one engine. Backed by Heavybit.

→ Automattic's interim CEO and its legal chief signed each other's severance deals while Matt Mullenweg was sidelined, and he fired both of them when he got back.

The board put Mullenweg on paid leave September 9. The next day, CFO and interim CEO Mark Davies and chief legal officer Andy Missan signed reciprocal agreements worth about $8.15 million combined, roughly $4 million each: a year of salary as a lump sum, accelerated vesting and a year of health coverage. Firing either for cause takes written notice within 60 days, a 30-day cure period and a board majority. Mullenweg was CEO again by September 12. The cure period was longer than the ouster.

→ Valor handed its investors $8.5 billion of SpaceX stock instead of cash, and its founder is keeping his.

About 8.5 percent of Valor's SpaceX position went out, so the whole stake was worth roughly $100 billion before the handout. Valor had put in a little over $400 million by 2021, which makes the distribution alone about 21 times everything it invested. LPs get shares and decide for themselves when to sell into a stock that is down since June. Antonio Gracias told Bloomberg: "I plan, personally, on holding my stake as long as I possibly can." That is an easy plan when the fund just sold nothing.

→ Rhodium says every Chinese AI model combined earns about a tenth of what OpenAI and Anthropic do.

The estimates published Thursday put OpenAI at $40 billion of ARR and Anthropic at $65 billion, against $4 billion for ByteDance, $2.4 billion for Alibaba, $1 billion for Moonshot, $800 million for MiniMax, and $500 million for DeepSeek. Z.ai told investors on Wednesday it now expects $3 billion by year end, up from $2.4 billion. The caveat is buried in the report: the figures reflect data available as of this summer, and ARR here means a recent month times twelve. Two of the most quoted numbers in AI are one month of revenue with a multiplication sign attached.

→ Anthropic and OpenAI say they will let outside safety testers work inside their labs, and neither has named a single tester.

Dario Amodei proposed embedded evaluators in an essay, and Sam Altman said OpenAI would commit too. TechCrunch lists what is missing: which evaluators, how many, when, what they can see and what they can publish. METR, Redwood Research, Apollo Research and FAR.AI are in the conversation, not on a contract. Meta, SpaceX AI and Google DeepMind have not signed on, so three of the five biggest labs are sitting out, and Demis Hassabis wants a standards body instead. For now the independent testers are four names in an article.

Still open: PearX W27, $500K to $2M per company, regular deadline October 4. pear.vc/accelerator

Always rolling: Thiel Fellowship, a $250,000 grant for young people who would rather build than sit in school. thielfellowship.org

Rhodium's whole report comes down to one test: who is getting paid. A Chinese lab with a 163x multiple and a startup with real revenue that nobody has heard of have opposite problems, and only the second one is fixable by us. Reply to this email and ask about Get Featured, and we will put your numbers in front of the founders and investors reading this.

Till next time,
The Vesting Team (Brycent, Ajay, and Kamil)