Anthropic's newly completed IPO on Thursday underscored a fundamental paradox in its business: despite a record $1.5 trillion valuation, the Claude developer still depends on Amazon and Google for a substantial share of its enterprise distribution and computing capacity.
Following the IPO, the two cloud giants now hold a combined stake of more than 28% in Anthropic, making them major shareholders, as well as infrastructure suppliers, distribution partners of Anthropic, whose models compete directly with their own frontier AI models.
Anthropic priced its shares at $83 on Wednesday, above its previously indicated range of $70 to $78. The offering raised approximately $81.3 billion, making it the largest IPO on record, surpassing the historic public offering of another competitor, Elon Musk’s SpaceX. Alongside the offering, Anthropic is selling $26 billion in non-voting Class N shares in a private placement to key cloud and chip partners, including Google, Microsoft and Nvidia.
For enterprise buyers, the IPO disclosures magnify the practical risks of relying on an AI vendor whose primary cloud distributors and compute suppliers are also rivals in the AI market.
Cloud deals: reach, and dependencies
Anthropic generated about $2.16 billion in sales on Amazon's and Google's cloud marketplaces in 2025, accounting for 47% of its revenue, according to the IPO filing.
These platforms let businesses buy third-party software using funds they’ve already committed to spending with cloud providers. Amazon and Google handle billing and payments for Claude purchases made on their platforms, while Anthropic also sells directly to customers.
At the end of 2025, marketplace sales accounted for 60% of the $909 million in Anthropic's outstanding accounts receivable, up from 42% in 2024. Because Amazon and Google handle those payments before transferring the proceeds to Anthropic, disputes or delays in collection could leave Anthropic waiting for cash it needs to meet its own expenses.
Rebecca Wettemann, CEO and principal analyst at Valoir, said the cloud giants give Anthropic a "front door to the enterprise" using existing procurement, security and billing systems.
But the relationships also create a potential conflict of interest. Amazon and Google are developing their own competing AI models and services while maintaining close commercial ties with Anthropic,
“These companies are simultaneously Anthropic's investors, infrastructure suppliers, distribution partners and competitors," said Matt Hasan, CEO of AIresults Global Advisory.
Hasan said Anthropic's long-term competitiveness depends in part on the companies that compete with it. Amazon and Google could favor their own models with pricing, deeper integration or more prominent positioning.
"Even small changes in computing allocations or sales priorities could put Claude at a competitive disadvantage," he said.
Amazon, however, can benefit from AI demand without relying solely on its own Nova models. CEO Andy Jassy has said customers want access to multiple leading models, allowing AWS to generate revenue by providing the cloud infrastructure to run third-party models such as Claude.
Computing needs deepen reliance on partners
The marketplace relationships are only one part of Anthropic's dependence on big tech. Its infrastructure commitments add another dimension. The IPO filing disclosed approximately $900.9 billion in multi-year committed spending, substantially all of which was for computing and infrastructure.
Named liabilities include at least $121.3 billion to Google and $110 billion to Amazon. Anthropic also committed to spending between $47.2 billion and $49.2 billion with Microsoft. These figures represent only some of Anthropic's infrastructure obligations.
Anthropic warned that some infrastructure projects depend on third-party financing and support to develop computing capacity. If financing becomes unavailable or more expensive, capacity could be delayed or fail to materialize. The company also cautioned that cloud providers could limit their computing allocations, even if contractual arrangements are in place.
Enterprises can inherit risks from these relationships even if Amazon and Google continue providing reliable infrastructure, said Jeet Pattanaik, founder and CTO of Glokal AI, which creates personalized AI systems for enterprises.
"If you rely on Claude, you're also relying, indirectly, on its cloud partners' capacity, pricing and priorities, and you don't have a contract with them for any of it," he said. "If one of those relationships shifts, customers feel the effects without having any say."
Hasan, of AIresults, said Anthropic's negotiating position depends on how much its partners need Claude and how much it can diversify its infrastructure.
"The more dependent it becomes on their computing capacity, the more bargaining power those partners could have," he said.
Cloud marketplace access no guarantee of easy exit
While buying Claude services from AWS or Google Cloud simplifies procurement and uses existing cloud commitments, it doesn't necessarily make switching providers easier. Companies can become over reliant on both the model provider and the cloud platform through which they access it.
Pattanaik said companies that build prompts, data pipelines and security settings around one cloud provider's AI services can face significant rework when switching. Teams that keep their prompts, tests and business logic separate from the model provider might switch in weeks, while those tied to one model's behavior could need months.
"A backup you've never tested isn't really a backup," he said.
Wettemann said CIOs should assess how easily they can substitute another model, compare costs across direct Anthropic access and cloud marketplaces, and understand what contractual protections cover availability, pricing and model transitions.
Big tech could gain more control
The longer-term risk for Anthropic is not that Amazon or Google will stop supporting Claude, but that their own AI models become good enough to reduce demand for Claude while the hyperscalers retain control over core infrastructure and enterprise distribution.
Wettemann cautioned that as models become more interchangeable, cloud providers could capture more value from their infrastructure and customer relationships. A July report on competition in AI markets from The Organisation for Economic Co-operation and Development also identified preferential pricing, product bundling and other advantages for vertically integrated companies as potential competition risks, especially when cloud providers develop their own AI models.
That could weaken Anthropic's bargaining power as its computing needs grow and its attachment to those partners deepens.
Hasan said the company's ability to maintain its competitive position will depend partly on diversifying its infrastructure and remaining valuable to its cloud partners.
"Anthropic's biggest competitive challenge may not be building better AI. It may be maintaining control over the economics of delivering it," Hasan said.
Kinza Yasar covers AI and emerging technology for TechTarget, with a focus on ethics, enterprise adoption, governance and business strategy. Before moving into journalism, she worked in IT and network support roles, giving her a systems-level perspective on how enterprise technologies are built, deployed and managed.