Anthropic Files S-1: $35B Trailing Revenue, 900% Net Retention

Anthropic filed to go public. The business is real: $35 billion in trailing revenue, positive operating cash flow, 900% net revenue retention. The filing is also the clearest look yet at what building on Claude actually rests on, and whether that ground shifts once public shareholders are in the room.

The retention number deserves skepticism. 900% net revenue retention sounds like a phenomenon. It might be ten accounts moving from $10M to $90M each, which is not the same thing as broad-based pricing power. The S-1 will disclose customer concentration in the risk factors. That disclosure tells you more than the retention headline. Similarly, 45% quarterly growth being described as a slowdown is silly in absolute terms, but the direction is real. The base is already enormous, and the math of a growing denominator always wins.

If you're a CTO negotiating a Claude contract, the timing matters. Anthropic has $131.7 billion in cash and every incentive to lock in logos before the listing. Post-IPO, quarterly margin pressure is a genuine force, and the first place it surfaces is in discounts that quietly stop renewing. Multi-year pricing, rate protection, and data handling terms are worth fighting for now. The leverage is better before the roadshow than after.

The safety question is structural. Anthropic sells patience on safety. Public markets do not reward patience on a quarterly clock. The Public Benefit Corporation structure gives the board some legal cover, but watch how the S-1 frames Constitutional AI and model evaluations. If they show up as business risks to be managed rather than commitments to be kept, that framing is informative about how the next conflict between safety timelines and revenue timelines gets resolved.

On compute, the cash pile is real but finite. $141.1 billion raised against $131.7 billion still in hand looks healthy until you account for forward compute commitments that don't sit on the balance sheet. A frontier training run at the next tier costs somewhere between $1 billion and $10 billion. Even this reserve buys only a handful of shots. Watch the roadshow for any signal on owned silicon versus continued dependence on hyperscaler GPUs. That ratio determines whether the margins ever accrue to Anthropic or to NVIDIA and the cloud.

The call: Anthropic's net revenue retention, as reported in its first full post-IPO quarterly results (Q4 2026 or equivalent), will come in below 900%. Medium confidence. The mechanism is straightforward: 900% is easiest to produce off a small early base where a few accounts scaling from pilot to production dominate the cohort. The filing already shows growth decelerating. As the installed base matures and early adopters approach their consumption ceiling, the expansion multiple on the full cohort has to compress. Holding at 900% would require the newest, largest customer cohort to expand faster than the early whales did, which runs against both the deceleration already visible and basic arithmetic.

Revisit by 2027-03-31. Right if post-IPO reported retention is under 900%. Wrong if it holds or rises.

Anthropic filed to go public, and the filing is doing two jobs at once. It tells you the business is real: $35 billion in revenue over the trailing year, positive operating cash flow, customers nine-ing their spend. It also tells you Anthropic is about to answer to public shareholders, which is a new kind of boss for a company that sells itself on taking the long view on safety. For anyone building on Claude, paying Anthropic bills, or depending on a second supplier to keep the first one honest, this filing is the clearest look yet at the ground under your feet.

What's actually being decided here isn't Anthropic's. It's yours. Do you keep routing more of your stack through Claude on the assumption that today's prices and today's reliability hold, or do you treat an IPO as the starting gun on a different set of incentives? That choice is easy to undo on the margin and expensive to undo once your agent pipelines are built around one provider's quirks. There's no hard deadline, but the IPO roadshow and the first two post-listing earnings calls will set the tone.

The Skeptic. 900% net retention is the number everyone will quote, and it's the one I trust least without the cohort behind it. Nine-times expansion is what you get when a handful of giant customers go from pilot to production. Ten accounts moving from $10M to $90M each produces the same headline as broad-based adoption, and it means something completely different for pricing power. The S-1 will disclose customer concentration somewhere in the risk factors. That clause tells you more than the retention figure. And the 45% quarterly growth being called a "slowdown" is silly in absolute terms, but it's real as a direction: the law of large numbers always wins, and $16.8 billion is a big denominator.

The Enterprise Buyer. If you're a CTO signing a Claude contract, an IPO changes what you can negotiate for and when. Right now Anthropic has every reason to lock in logos and let spend expand. $131.7 billion in cash means they can keep prices friendly to feed that retention story through the listing. Post-IPO, quarterly margin pressure is a real force, and the first place it shows up is in the discounts that quietly stop renewing. The move is to get multi-year pricing and rate protection in writing now, plus data handling and indemnification terms that survive a change in ownership incentives. The leverage is better before the roadshow than after.

The Safety Lens. Anthropic sells itself as the lab that will slow down when safety demands it. Public markets don't reward slowing down on a quarterly clock. The Public Benefit Corporation structure gives the board some cover, but fiduciary duty to shareholders pulls the other way every time a safety-motivated delay costs a quarter of revenue. Read how the S-1 frames its evaluation and alignment work. If Constitutional AI and model evals show up as business risks to be managed rather than commitments to be kept, that's your signal about how the board resolves the next conflict. $131.7 billion also buys capability jumps faster than anyone's ability to evaluate them.

The Compute Pragmatist. The cash pile is real but it's not idle. $141.1 billion raised against $131.7 billion still in hand looks like modest burn until you remember the raises are staged against forward compute commitments that don't sit on the balance sheet. Positive operating cash flow means inference pays the bills. Training is still a cost center, and a frontier training run at the next tier runs somewhere between $1 billion and $10 billion. Even this hoard buys a handful of shots before the next raise. Watch the roadshow for capex guidance and any hint of owned silicon versus staying renters on hyperscaler GPUs. That ratio decides whether the margins ever belong to Anthropic or to NVIDIA and the cloud.

Where they disagree

The real split is between the Enterprise Buyer and the Skeptic, and it's about who has the leverage. The Buyer assumes Anthropic needs your logo and will deal to keep it. The Skeptic suspects the revenue is concentrated enough that Anthropic doesn't need your logo, it needs the ten whales, and everyone else is a rounding error with no negotiating power. Both can't be right. If retention is broad, mid-market buyers can push for terms. If it's ten accounts, Anthropic sets the price and you take it.

The second tension is Safety versus Compute. The Safety Lens worries public pressure forces Anthropic to ship faster than it can evaluate. The Compute Pragmatist points out the cash only buys a few frontier runs, which is its own brake. Scarcity of compute shots might enforce more caution than any governance structure.

What it hinges on

Two facts settle most of this, and both are in the document. First, customer concentration: is the 900% built on a broad base or a short list of hyperscalers and labs? That determines whether anyone outside the whale list has pricing leverage. Second, the trajectory of that 45% quarterly growth: one more quarter tells you whether this is healthy maturation or a wall. Before you deepen your commitment to Claude, pull the concentration disclosure and price your own contract as if the discounts stop renewing after the IPO. Build and test a fallback path to a second model now, while switching is cheap.

Prediction: Anthropic's net revenue retention, as reported in its first full quarterly results after the IPO (the Q4 2026 or first post-listing earnings report), will come in below the 900% figure in the S-1.

Confidence: Medium. The math of a growing base forces the ratio down.

Why: 900% retention means last year's customers spend nine times as much today, and that multiple is easiest to hit off a small early base where a few accounts scaling from pilot to production dominate the number. The filing already shows quarterly growth decelerating to 45%, which is what happens as the denominator gets huge. As the installed base matures and the big early adopters approach the ceiling of what they can consume, the expansion multiple on the full cohort has to compress, because you cannot 9x a base that's already enormous. The opposite outcome, retention holding or rising, would require the newest and largest cohort of customers to expand even faster than the early whales did, which runs against both the growth slowdown already in the filing and the plain arithmetic of a bigger starting number.

Revisit by 2027-03-31: We're right if Anthropic's first post-IPO reported net revenue retention is under 900%. We're wrong if it is 900% or higher.