What enterprises can learn from Anthropic’s IPO prospectus about vendor dependency

Anthropic is set to go public in late November at a valuation of between $1.5 trillion and $2 trillion. The higher end of the range would make it the biggest initial public offering (IPO) ever, beating SpaceX’s $1.8 trillion public debut in June. The Claude creator filed its much-awaited IPO prospectus on Monday 26 October, giving investors and tech leaders a peek into its financials, especially its dependencies.

The document, also known as an S-1, highlights a reliance on a limited number of customers for its revenue. Nearly half (47%) of sales in 2025 were routed through its cloud partners, Amazon and Alphabet’s Google. Meanwhile, each of its two biggest customers, who remained unnamed, accounted for 12% of revenue last year.

Some progress is being made to reduce this dependence. While 48% of revenue between January and September this year came through its cloud providers, the five biggest customers accounted for approximately 25% of revenue for the nine-month period. No single customer had a share greater than 10%.

The company noted in its filing: “We anticipate that we will continue to derive a significant portion of our revenue from a limited number of customers for the foreseeable future, and in some cases, the portion of our revenue attributable to certain customers may increase in the future.”

The IPO prospectus has raised questions about an overreliance on AI models and vendors – and the risks this brings for enterprises.

“It’s a reminder that the model, the vendor and the route you access it through can all change for reasons outside your control,” cautions Jan Karstens, CTO at AI-powered SAP automation platform Avantra. For example, enterprises that have built their critical business processes around one AI supplier could face operational disruption, even if it’s temporary.

“We’ve already seen what that looks like. When Anthropic was ordered to block foreign nationals from its top models this summer, the only way to comply in time was to switch them off for everyone. Teams that had already assumed any model or vendor could disappear overnight barely noticed. Everyone else had to scramble,” says Karstens.

Charles Radclyffe, CEO of SaaS enterprise provider EA Global AI adds: “The interesting thing about the reported Anthropic numbers is that they're not just a story about one AI company. They're a reminder that concentration risk runs through the AI ecosystem, and enterprises need to understand where those dependencies sit.”

Earlier this year, automation platform Zapier polled 500 US executives and found that 47% of enterprises would stop functioning properly if their primary AI vendor were to shut off its service tomorrow. Just over a quarter (27%) indicated they were reliant on AI vendors for nearly all of their business operations.

Question vendors’ own dependencies

Part of the problem is that enterprises focus too much on model performance, when they should be questioning their AI service’s supply chain.

“If a provider depends heavily on a small number of cloud platforms, strategic partners or major customers, changes in those relationships can ultimately affect enterprise users through pricing, availability, product direction, or service levels,” says Radclyffe.

So, what does he think CIOs should be doing? Asking companies supplying the AI services about their own critical dependencies, how workloads are delivered, the knock-on effect if key commercial relationships change, and what notice customers receive before major model changes. “AI procurement needs to become a resilience conversation,” he stresses.

Karstens believes asking model makers about their critical dependencies is reasonable, but he cautions against acting on their responses alone. He’d “be mildly nervous about concentration among my direct AI vendors. I’d answer that with my own visibility and multiple suppliers, not with questionnaires.”

CIOs need to “start from the assumption that any model, vendor or access path can be withdrawn without warning,” he adds. “They should build governance that doesn’t depend on any one of them staying in place. Periodic audits and vendor assurances after the fact aren’t enough when the [AI] market moves fast.”

Be wary about vendor lock-in

A key takeaway from Anthropic’s S-1 is that consumption-based, or per-token, revenue made up 83% of revenue in the first six months of 2026. However, many of its main customers are not locked into long-term contracts.

AI model makers are pushing hard for customers to agree to contracts to secure multi-year revenue visibility amid fierce competition. If enterprises find themselves signing contracts, then they need to ensure they agree to “meaningful exit rights, portability provisions, and the ability to export not just data but the prompts, workflows and evaluations that support business processes,” advises Radclyffe.

Karstens agrees. “Vendors like Anthropic are moving deeper into everyday work, with documents, slides and projects now living inside the AI tool itself. If that continues, data portability will become a real issue for CIOs in the next 12 to 24 months.”

The Zapier study also surveyed senior executives about AI vendor lock-in. Just over eight in 10 (81%) admitted they were at least slightly nervous about being locked into contracts, with 29% very concerned. Almost half (46%) were worried about data migration and an overreliance on a single vendor. Sudden price hikes were a concern for 41%, while another major worry for 37% was having a single point of failure.

Ideally, enterprises should avoid finding themselves in a situation where these worries come to fruition by ensuring they have a practical alternative to turn to if costs rise, there’s a change in circumstances, or unexpected events.

Radclyffe sums it up perfectly: “There's nothing unusual about [vendor] concentration in a fast-growing technology market. The risk comes when enterprises don't understand how dependent they've become until they're forced to make a change.”