Anthropic’s Leaked S-1: $16.3B in Six Months, an $80.7B Loss, and Two Customers Worth a Fifth of It

Introduction: The Draft Prospectus Goes Public

The Wall Street Journal got a look at Anthropic’s draft S-1 prospectus on October 20, 2026, and for the first time the company behind Claude has hard, auditable numbers attached to the mega-IPO it has been teasing since the summer. The company declined to comment on the report, but the draft tells the story anyway: revenue is compounding faster than almost any company in history, the paper losses are staggering but mostly an accounting artifact, and the risk factors read like a list of everything the AI tools market has spent 2026 arguing about.

We covered the $2 trillion valuation chatter in August, when the listing was still rumor-shaped. Now there are real filings — and if you run any part of your workflow on Claude, or on tools built atop it, the details are worth ten minutes of your attention.

The Numbers: 15x Growth and an $80.7 Billion “Loss”

The headline figures from the draft prospectus, as reported by the Journal:

• $16.3 billion in first-half 2026 revenue — up from about $1.1 billion in the same period last year, a nearly fifteenfold jump, and already more than triple the roughly $4.6 billion Anthropic earned in all of 2025.
• An $80.7 billion net loss for the half — but roughly $72 billion of that comes from accounting charges tied to convertible notes and other financing instruments, not from operations. The prior-year half-year loss was about $13 billion.
• Nearly $9 billion in compute and infrastructure spending in six months — more than all of 2025 (~$7.3 billion), and up from about $2.3 billion a year earlier.
• $16.8 billion in Q3 revenue, up 45% quarter-over-quarter — a figure Reuters attributed to an investor who attended a pre-IPO meeting, meaning the growth didn’t stall after June.

The important nuance: the eye-watering loss is dominated by non-cash charges from how Anthropic’s late-stage financing rounds convert into equity. Cash burn is real — compute spend alone nearly quadrupled year-over-year — but the $80.7 billion figure is not $80.7 billion of lost operating money. Investors will price both separately; so should you.

The Risk Flag: Two Customers, One Fifth of Revenue

Here is the line that will dominate IPO coverage: roughly a fifth of Anthropic’s first-half revenue came from just two customers, and the prospectus reportedly warns that a limited number of clients will keep generating a significant share of revenue “for the foreseeable future.” Neither customer is named, and per the Journal, the deals are not bound by long-term contracts.

For context, those two accounts likely represent some mix of cloud resellers, coding-tool platforms, and enterprise deployments — the same layer of the stack where Claude Code, Cursor, and GitHub Copilot compete to route your subscription dollars to a model vendor. It’s the same dynamic we saw when Replit switched its default from Claude to OpenAI this week: platform defaults move billions in revenue, they move fast, and they are renegotiable.

That cuts both ways for tool buyers. The concentration means Anthropic’s growth depends on a handful of platforms keeping Claude as a default — which gives big partners enormous pricing leverage, and gives Anthropic every incentive to keep winning developers with better models and sharper tooling. Public shareholders, though, will be watching that top-two number every quarter.

$600 Billion in Compute Commitments — and Counting

The draft prospectus reportedly shows infrastructure commitments exceeding $600 billion, with additional deals under negotiation that could add up to $150 billion more. That is the largest compute lock-in ever committed to paper, and it lines up with the rest of 2026’s infrastructure arms race: the SpaceX deal we covered in September, and — reported the same day as the S-1 leak — Bloomberg’s story that Anthropic and UK-based Nscale are in talks for a second ~$45 billion West Virginia compute deal, on top of the 460 MW, six-year Monarch Compute Campus agreement reported in August.

Why should an AI tools buyer care about data-center contracts? Because commitments this size are promises to spend regardless of what revenue does. If growth slows, the gap gets closed somewhere — and pricing on APIs, subscriptions, and the model tiers inside your favorite tools are the obvious levers. It’s the compute-crunch story we’ve documented all year, now with a prospectus-sized number attached.

What a Public Anthropic Means for Your AI Stack

You don’t have to have an opinion on the IPO to position yourself well. A few practical reads of the filing:

• Expect pricing volatility, in both directions. A listed Anthropic must defend the revenue growth the valuation assumes, which argues for aggressive pricing and free tiers to win volume — but margin discipline could push flagship-model prices up. Build for movement, not stability.
• Avoid single-vendor tooling. The concentration-risk disclosure is the strongest argument yet for multi-model stacks. Tools that let you route between Claude, Gemini, DeepSeek, and open-weight models insulate you from any one vendor’s quarter-end decisions.
• Watch your agentic burn. Autonomous coding agents consume tokens without asking. If post-IPO pricing shifts, agent-driven spend feels it first — meter it now.
• Read the risk factors as a checklist. Customer concentration, compute commitments, AI backlash: everything listed in the S-1 is a proxy for where public pressure will push the company after listing.
• Don’t panic about the loss. Accounting charges aside, the operating picture (revenue up 15x, Q3 accelerating) is why this listing is credible at all. The real question for users is capacity: $600 billion of compute is being bought to serve demand that tools like yours generate.

Frequently Asked Questions