Anthropic Filed Its S-1. The Numbers Inside Are the Whole Story.

On Tuesday morning, Anthropic's AI went dark for everyone on earth, The Independent confirmed errors across Claude's models, with service restored by noon Eastern time. It was the world's most valuable AI company, dependent on three other companies' data centers — and for several hours, it had nothing to offer its users.

The outage lasted less than a day. But it exposed the central paradox of the AI industry's current moment: the companies valued highest are often the most structurally dependent on someone else's infrastructure. Anthropic has never owned a server. Its entire compute backbone runs through Amazon, Google, and SpaceX. It is, by any conventional measure, a model maker and enterprise software company — not an infrastructure business.

Anthropic closed a $65 billion Series H funding round on May 28, valuing it at $965 billion post-money. Revenue run rate crossed $47 billion earlier this month, according to the company's announcement. For context: no company in any sector has grown from $9 billion to $30 billion in annualized revenue in four months. Not OpenAI. Not Stripe. Not any hypergrowth SaaS company on record, according to MindStudio analysis.

Anthropic has also confidentially submitted a draft S-1 registration statement to the SEC, the company said in its announcement — filed under JOBS Act Section 6(e), which allows emerging growth companies to keep financial disclosures confidential until 21 days before a roadshow or public listing. The actual cost structure, margin assumptions, and revenue composition in that filing remain hidden from public view. An SEC EDGAR search for confidential S-1 submissions does not surface the document at this time; the filing will become public either ahead of a roadshow or on request. What is known from the public announcement: the company has disclosed compute agreements with Amazon for up to five gigawatts of new capacity, with Google and Broadcom for five gigawatts of next-generation TPU capacity, and with SpaceX for access to GPU capacity in its Colossus supercomputer. The numbers that matter — what Anthropic actually pays for that compute, and what margins it keeps — are in the sealed filing.

The valuation raises immediate questions for public market investors. At $965 billion post-money against $47 billion in annualized revenue, Anthropic trades at roughly 20 times forward revenue. For context, OpenAI's most recent disclosed valuation implies a similar multiple. Frontier AI infrastructure companies that sell compute access — NVIDIA trades at roughly 30 to 35 times forward revenue — carry premium valuations because they own the underlying resource. Anthropic does not own the compute it sells. If Anthropic reaches Rule of 40 metrics — a common SaaS health benchmark where revenue growth rate plus profit margin exceeds 40 percent — the valuation may be defensible. If margins compress as hyperscalers extract more value from their structural position, the math becomes difficult. The $965 billion number was set between investors who wanted in and a company that could name its price. The public markets will ask a different question: what do you own, what do you control, and what do you pay for the infrastructure you depend on?

The competitive picture supports the valuation in the near term. Anthropic holds 42 to 54 percent of the enterprise coding segment of generative AI, compared to OpenAI's 21 percent. Claude grew from 1 percent of global AI app downloads in the second quarter of last year to 14 percent in the same period this year. ChatGPT declined from 67 percent to 47 percent over the same span. In the two weeks following OpenAI's announcement of a Pentagon partnership, ChatGPT uninstalls surged 295 percent day-over-day, and Claude hit number one in the App Store. The safety positioning is converting.

Independent analysts are less uniform. Futurum Group notes that while 78 percent of organizations expect to increase AI budgets in the next year, 63 percent still allocate 10 percent or less of their total tech budget to AI. Reliability and hallucination management remain the top adoption challenge for 55 percent of buyers; data privacy for 53 percent.

The compute agreements are the strongest argument for the valuation's durability. Amazon Web Services, Google Cloud, and Microsoft Azure all now offer Claude as a first-party service. Anthropic has locked up access to some of the largest compute infrastructure in the world. Three cloud providers, none of whom can afford to let Claude fail. But the dependency runs both directions. Anthropic's margins are a function of what it pays Amazon, Google, and SpaceX for the chips that run its models. If those costs rise — as hyperscalers gain pricing power over a client that has no alternative — the path from $47 billion run rate to actual profitability narrows.

The filing is with the SEC. The numbers that matter are still sealed. The public markets will provide the answer.