Anthropic's IPO is expected later this month, and the story it is telling buyers is not the one it started the year with. The old pitch was Claude Code's growth. The new pitch is a chip story.
Growth Built the Case, and It Bent in Q3
Claude Code became the coding tool enterprises bought. More than 11,900 customers now pay Anthropic $100,000 or more a year, and OpenAI's Codex fell behind in enterprise sales earlier this year. In Q2 the company posted its first operating profit, $974 million.
Then the amended S-1 landed on November 2, and it showed growth slowing. Revenue grew 144% from Q1 to Q2, then 47% from Q2 to Q3. Still fast, but a third of the pace. Profit flipped too. Anthropic swung to a $567 million operating loss in Q3, mostly because it had to buy $5.0 billion of compute on short-term deals, at a premium.
Anthropic's revenue, revenue per $1 of compute (revenue divided by compute and infrastructure expense), and operating result, Q2 vs Q3 2026. The Q3 loss includes a $2.0 billion non-cash charge. Graphic: EO, from Anthropic's S-1 and S-1/A.
In two weeks, three Claude Code people left, including product lead Cat Wu. Wu went straight to OpenAI, reporting to the executive who has led its Codex work.
Weeks before a listing, one of the people behind Anthropic's growth engine has walked over to its closest rival. While that looks like a red flag, Anthropic’s pitch to IPO buyers has already moved on from the growth of Claude Code.
Anthropic Is Pitching Chips to Fix Its Compute Bill
Compute is what turned Q3's profit into a loss. In its investor meetings, Anthropic pitched a way to cut that bill. A new 15-person chip-design team, it told prospective buyers, already uses Claude for chip engineering. The aim is cheaper chips built for Anthropic's own models, and less reliance on capacity it rents from others.
The pitch builds on something Anthropic says is already happening in software. Claude wrote over 90% of contributions to its own code base in September, by the company's own count. An outside evaluator backs part of the claim. METR (an AI evaluation nonprofit) says Anthropic's models are improving about twice as fast as they did from early 2024 to late 2025, mostly because of Anthropic's own use of AI.
Anthropic carries about $895 billion in compute purchase commitments. A bill that size needs a margin story to stand behind it.
Tomasz TunguzGeneral Partner, Theory Ventures
It is 15 people, and Anthropic still rents Trainium, TPU, and Nvidia capacity. METR also says Anthropic redacted most of the evidence it found most informative. Today this is a pitch, not a chip.
The Window Is Open for One Company, and It Is Narrowing
The cover page of Anthropic's Form S-1, filed with the SEC on October 26, 2026.Source: SEC EDGAR.
Anthropic has the IPO market nearly to itself. No significant AI company has listed since July, and OpenAI isn't expected to try until 2027.
The price is already sliding. Valuation talk has fallen from about $2 trillion to $1.5 trillion, and three major investors who plan to buy in told the WSJ it will likely need to come down further. Another wondered whether Anthropic might delay.
On November 3, Democrats swept Congress after running against Big Tech, and voters rejected data centers in 44 of 45 ballot measures. Two days later, the Nasdaq Composite had its worst day since April 2025. That matters to a company whose $895 billion in commitments assumes data centers get built on time.
The last big AI listing shows what public buyers do. AI chipmaker Cerebras cut its margin forecast at its first earnings and fell below its IPO price.
Compute ate 48% of revenue in Q2 and 58% in Q3. The first public quarters could show whether that share falls back. That is the number the chip story promises to move.
With the midterms' blue wave already darkening the outlook for AI, Anthropic's IPO and its price will set the reference point for every late-stage AI round. If the chip story holds, the rest of the pipeline follows it out. If it doesn't, every late-stage round priced off Anthropic comes down with it.
