In the familiar story of platform companies, an empire rarely looks like an empire in its first filing. What looks like a modest product in one decade can become the architecture of an economy in the next. Investors who read a filing literally see a product. Investors who read it as an architecture book see an empire. The difference was knowing that the real product was never the merchandise. It was lock-in.
Anthropic's newly public S-1 filing describes that arc in plainer terms. "Our customers' journey using Claude often begins with a single use case before expanding across departments, applications, and workflows," Anthropic said in its prospectus, which CNBC reported alongside a net revenue retention rate of around 900% as of Sept. 30. Existing customers are increasing their spending ninefold from one year to the next. Whether or not that feels familiar is an analysis, and it matters to operators far more than the hype around model quality ever does.
The Amazon playbook, distilled
Amazon's genius was never pricing or logistics alone. It was the discipline of making each new product lower the customer's incentive to leave the last one.
The mechanism is simple and relentless. Each extra product a business adopts raises the cost of leaving, not in any single invoice, but in accumulated friction. Teams build on it. Data lives on it. Risk concentrates around it. A company may be paying less at each point than it would at a competitor, yet by the time it considers switching, the switching cost has become a wall. The vendor does not win on any one deal. It wins on the combined lifetime value of a customer who can no longer easily leave.
What the filing actually describes
The filing shows how far this expansion has already gone. As of the end of September, Anthropic had more than 11,900 customers spending $100,000 or more annually, 1,900 customers spending $1 million or more, and about 250 customers spending $10 million or more. Growth has been driven largely by the popularity of the AI coding tool Claude Code and newer models including Claude Fable 5 and Mythos 5, CNBC reported.
What the filing does measure is retention, and the number is striking. Net revenue retention stood at around 900% as of Sept. 30, according to the prospectus, versus 108% for the median public cloud software company as of January. To an investor, that reads as a sign that customers who arrive for one use case keep deepening the relationship. To a buyer, it should read as a signal that the vendor's incentives are built for exactly that deepening.
What this means for the buyer
For an operator, the immediate lesson is that the deal structure is working against your impulse to say yes quickly. The vendor is structurally rewarded for you saying yes to the next thing, because each yes makes you a little harder to replace and a little more valuable to retain. The rational response is not to refuse adoption. It is to adopt with your options fully visible.
Practically, this means keeping data portable, wrapping models behind a layer you control, and treating a single AI provider as one option in a portfolio rather than a foundation. The companies that get trapped are the ones that optimize for speed and let lock-in become an accident. The companies that come out ahead optimize for reversibility, and let stickiness be a choice they keep making only because the product earns it.
The leverage that stays with the customer
None of this means the customer is powerless. Lock-in is conditional, and the vendor knows it. Anthropic, like Amazon, still needs new customers, still needs to defend against competitors, and still cannot treat any customer as captive without paying the price in churn, pricing pressure, or lost deals. The buyer's leverage lives in the simple but underused fact that switching is still possible. A vendor that believes it has fully locked you in has already priced that belief into its ambition, and ambition is a liability when the competitor makes a better offer.
This is why, in my view, the strongest operators should be quietly building their own portability standards, multi-tenant and multi-provider architectures, and internal platforms that make it trivial to rebalance work across vendors. The moment a customer can move a workload in days instead of months, the vendor's expansion stops being a trap and becomes a marketplace. The leverage flips back to whoever controls the exit.
Who really wins the deal
The honest answer is that the vendor and the customer win on different timelines, and the question only resolves if you specify the horizon. The vendor wins the short game. A playbook built on expansion from a single use case raises the combined lifetime value of a customer precisely by making each adoption sensible and each departure costly. That is a strong business model for the company selling it, and the ninefold spending growth in the filing is the clearest evidence that it is working.
The customer wins the long game only to the extent it refuses to confuse convenience with dependency. Whoever keeps the exit real, the data portable, and the portfolio open retains the power to convert the vendor's stickiness into a negotiating asset rather than a trap. The vendor wins the deal on the invoice. The customer wins the relationship only by never letting the invoice be the whole story.
So who is winning? The vendor who designed the expansion structure is winning the deal as it closes. The customer that understood the structure while signing is winning the next one. In the end, whoever controls the ability to leave controls the real deal, and stickiness only rewards those who build it on purpose.
Updated October 30, 2026: We reviewed this article against its sources and corrected statements they did not support.
Sources
- Anthropic files S-1 with the SEC for proposed initial public offering (Anthropic, 2026-10-26)
- Anthropic IPO live updates: AI lab's S-1 filing (CNBC, 2026-10-26)
- Anthropic officially files for its public debut (TechCrunch, 2026-10-26)
