AI & ComputeLOT 261122 · 4 min read

Anthropic's IPO Pops to $111, but $900 Billion in Compute Bills Loom

The first pure AI lab to list priced at $83 and touched $111 on Nasdaq. Ahead: a 17-fold revenue climb, $900B in compute commitments and a governance maze built to keep founders in control.

By Rebecca Stone798 words

Key takeaways

  1. Anthropic priced its Nasdaq IPO at $83 a share and traded up to $111, the first pure AI lab to go public.
  2. Compute commitments total roughly $900 billion over the coming years, per Anthropic's S-1.
  3. Seven founders control 50.1% of voting power through a single share held by Founder LLC.
  4. Nvidia agreed to buy $10 billion in Anthropic shares via private placement, on top of $10 billion promised last year.
  5. Analyst Steve Rattner estimates Anthropic must grow revenue roughly 17-fold in 10 years to match competitor returns.
Can Anthropic afford to put safety first?
Can Anthropic afford to put safety first? — Pexels / pixabay

Anthropic priced its IPO at $83 a share last Thursday and the stock ran to $111, making the Claude maker one of the most valuable companies on the planet and the first pure AI lab to list on the Nasdaq. The numbers stunned the market. They also bought the company a deadline: roughly $900 billion in compute commitments over the coming years, according to its own S-1 filings, that revenue now has to cover quarter after quarter.

The Verge's senior AI reporter Hayden Field walked through the economics on the Decoder podcast, and her read is blunt. "I saw one estimate from Steve Rattner saying Anthropic is going to have to increase its revenue roughly 17-fold within the next 10 years to deliver a comparable return to some of its competitors in the tech industry," she said.

What does the IPO actually prove?

Not much about the AI bubble, in Field's view. "This is peak AI hype right here," she said, noting the listing was on track to break records for months and ultimately beat Elon Musk's benchmarks. She's heard the IPO will mint thousands of new millionaires and hundreds of centimillionaires.

The hype has a ceiling. Field chatted with tech workers across several companies who say they're being pushed to do more AI even where it makes work worse — a reckoning, she argues, is coming. Anthropic also beat OpenAI to the exit, and OpenAI seems content about it. Field heard Altman was once obsessed with listing first, but after a year of scandals and public backlash, OpenAI stepped back and let Anthropic carry the trial balloon.

Can a token sell for more than it costs?

The business, stripped bare: Anthropic rents GPU capacity from hyperscalers, pays for power and chips, and must sell each token for more than those inputs cost. Field sees pressure from both ends.

Microsoft CEO Satya Nadella, speaking at the Ignite conference, said he wants token prices to fall and will route Copilot traffic to the cheapest capable model. Anthropic needs the opposite — fatter margins per token to fund training. Last year it faced a class-action backlash over subscription price hikes, rolled some back, and now leans on two levers: locking customers into a moat of integrated tools, and hoping enterprise and government contracts paper over consumer price sensitivity.

There's friction even there. Big customers demand data sovereignty, which kills monetization options. Anthropic's answer, Field argues, is the safety brand itself. Small and mid-sized businesses are building internal benchmarks to route only ~10% of tasks to expensive Anthropic models, sending the rest to open-source or mid-tier alternatives.

Who really controls the company?

The governance structure is where the story darkens. Seven Anthropic founders hold a single share of stock — issued to an entity called Founder LLC — carrying 50.1% of the voting power.

Above them sits the Long-Term Benefit Trust, a body of four people with little AI experience among them: former Federal Reserve Chairman Ben Bernanke, foreign policy analyst Richard Fontaine, experimental physicist Christopher Stubbs, and former GiveWell executive Neil Buddy Shah, who chairs it. The trust can appoint or remove certain board members, but only with 30 days' notice to a committee that includes co-founder Daniela Amodei. Of seven board seats, the trust picks three, founders pick three, and the tie-breaking seventh requires what amounts to founder approval.

One former employee told Field the trust was handpicked not to rock the boat. Unlike even OpenAI's old nonprofit board, it cannot block a model release. Law professor Ann Lipton of the University of Colorado told Field the real constraint isn't governance at all — it's that Anthropic needs money.

What about the Nvidia entanglement?

Dario Amodei told JPMorgan Chase's CEO in a livestream that compute is the company's "deepest business dilemma": buy too much and you overextend financially, buy too little and you can't serve inference or fund R&D. Anthropic underinvested in compute relative to OpenAI, Field notes, and is now paying for it.

The Nvidia relationship deepens the knot. Nvidia agreed to buy $10 billion in Anthropic shares via private placement, on top of $10 billion in funding promised last year. With Jensen Huang out after a board-driven resignation over conduct violations, Field wonders whether Nvidia's new CEO can still wheel and deal — including on diverting China-bound chips back to the US supply pool, which could ease Anthropic's constraint.

The watch item going forward, per Field: the earnings reports and what gets quietly sidelined. "The company that made its bed on safe AI is increasingly looking the same as OpenAI," she said, predicting "swept-under-the-rug" updates as the pressure to be first compounds.

theverge.com (Source)

Filed under

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering technology and product news in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,245 articles

Read also

« Previous article