Anthropic's IPO filing shows cross-selling is what keeps enterprise Claude customers locked in

The number that decides the renewal is 900%

Anthropic's Form S-1, filed with the SEC on Monday, puts net revenue retention at about 900% as of the end of the third quarter. Net revenue retention measures what the existing customer base spends this period against what the same base spent last period, so a figure near 900% means the average account multiplied its spend roughly ninefold inside a year without a new logo being signed.

That is the number to carry into a vendor call, because it describes the trajectory of the contract rather than its current price. Anthropic credits it to what the filing calls significant usage expansion as customers deploy Claude across their organisations. A buyer negotiating a multi-year term is not pricing today's consumption; they are pricing the consumption the vendor's own filing says tends to follow.

The account counts behind the percentage are in the same document. As of September 30, Anthropic recorded more than 11,900 customers spending at least $100,000 a year, more than 1,900 spending upward of $1 million and about 250 spending $10 million or more annually. Those are the accounts where a renewal conversation has a procurement lead in the room.

Cross-selling is the mechanism, and the filing says so

More than half of customers spending at least $100,000 annually now use both the Claude platform and a first-party app like Claude Code or Cowork, up from about 30% at the end of last year. That shift, from roughly three in ten six-figure accounts to more than half, is the clearest evidence in the document that the bundle is doing the work.

Anthropic describes Claude as three layers: the models, the developer platform and the applications users interact with directly. The filing argues that an advance at any layer compounds across all three, and calls this integrated approach the reason more customers use both the platform and first-party apps. Whether or not a buyer accepts the compounding claim, the commercial consequence is straightforward: a business that starts with the API can add Claude Code for its engineers and then Cowork for the wider workforce without a new procurement cycle.

The filing names the comparison it wants. TechCrunch reads the strategy as the same cross-product ecosystem that made Amazon and Uber sticky, and the S-1 itself pitches Claude as not a static product that serves a single use case but frontier AI with rapidly advancing capabilities. For a buyer, that framing matters at renewal: the vendor is selling a platform whose value is claimed to grow with each layer added, which is also an argument for adding the next layer rather than renegotiating the current one.

Where the spend actually grows: the 20x coding gap

In the third quarter, the highest-spending 5% of Claude Enterprise customers who used Claude Code spent about 20x as much per month as the average enterprise user who did not use the coding tool, per the filing. The comparison is between the top spenders who adopted the tool and the average user who did not, so it is not a like-for-like uplift on a single account. It is still the number that explains why the applications matter to the vendor's revenue line.

The practical reading for an IT director is that engineering adoption is the line item that moves. A coding assistant seats itself with the people most able to generate usage, and usage is what a consumption-based platform bills for. An estate that pilots the API with a small team and then rolls Claude Code to its engineering organisation should expect the monthly figure to be set by that second step, not the first.

Anthropic says it amplifies the diffusion deliberately. An internal Applied AI team works with startups and enterprises to place Claude inside existing technology stacks, and the company recently launched a venture with Blackstone and other private equity firms, called Ode with Anthropic, to embed forward-deployed engineers into businesses building bespoke AI tools with Claude. Those teams lower the cost of adding a layer, which is the same thing as raising the cost of leaving one behind.

The distribution deal that keeps the contract with Anthropic

Claude reaches enterprises through AWS, Google Cloud and Microsoft Azure under what the filing calls a joint go-to-market approach. The structure is worth reading closely: Anthropic holds the direct contractual relationship with the customer and pays each cloud provider a fee, which the company says lets it avoid channel conflict and set pricing.

For a buyer, that means the hyperscaler is a route to the product, not the counterparty on the paper. Committed cloud spend drawn down against Claude does not necessarily make the cloud provider the party you negotiate with, and the pricing authority the filing claims sits with Anthropic. Procurement should establish which entity signs before assuming an existing cloud agreement carries any leverage.

The same partnerships are listed among the risks. Anthropic says the cloud providers that help sell Claude are increasingly seeing it as a competitor, which could make it harder to secure the compute it needs from them, and the filing states Amazon, Google and Microsoft may prioritise their own requirements or rivals they have invested in. A buyer standardising on Claude through one hyperscaler is therefore exposed to a commercial relationship the vendor itself flags as shifting.

What the filing admits about concentration

The risk factors name customer concentration directly. The filing warns that any adverse change in demand from its largest customers could adversely affect its business and prospects, and that a significant reduction in spending by, or the loss of, one or more of its largest customers, if not offset by new relationships or increased business from existing customers, could adversely affect its financial condition and results of operations.

That cuts both ways for an enterprise buyer. A vendor with about 250 accounts above $10 million a year has strong reason to keep them, which is negotiating leverage in the reader's favour. The same concentration means the roadmap and the commercial terms are shaped around a small set of very large customers whose requirements may not resemble a 2,000-seat estate's.

The filing also says the company will devote less effort than a typical business to entrenching its current products and business models and more to defining new ones, aiming at new classes of products whose aim is to dwarf the economic value of existing ones. A buyer signing a three-year term should ask what happens to the layer they standardised on when the vendor's stated priority is replacing it.

The decision rule for the next AI vendor review

Choose the bundled route if the estate will adopt more than one layer within the term. The filing's own evidence is that accounts crossing from platform-only to platform plus application move from about 30% to more than half of the six-figure base, and that the coding tool correlates with the heaviest spend. If engineering adoption is already planned, negotiate the second layer into the first contract rather than buying it at list later.

Choose a narrower commitment if the requirement is a single workload. The published numbers describe a vendor whose revenue model rewards expansion, with net revenue retention near 900% and a top cohort spending about 20x the average non-coding user per month. A buyer who cannot forecast that expansion should cap consumption, fix a term and ask for the per-unit rates, which the filing does not state.

Whichever route, ask three questions before signing. Which Anthropic entity is the counterparty when the deal arrives through AWS, Google Cloud or Azure; what the per-seat and per-unit rates are, since the filing publishes none; and what commitment, if any, survives if the vendor's stated plan to reinvent its product line retires the layer being bought. The filing answers the strategy question and leaves the pricing ones open, which is exactly what the renewal meeting is for.