Anthropic is reportedly struggling to secure enough compute, the processing power used to train and run AI models, from cloud providers that increasingly see it as a competitor. Its infrastructure spending is ballooning at the same time, ahead of a planned November IPO.
The tension is built into how Anthropic got here. Amazon and Google are both investors in the company and major suppliers of the chips behind Claude, Amazon through its Trainium processors and Google through its TPUs. In October 2025, Anthropic announced a deal with Google for up to 1 million TPUs. A month later it committed to $30B of capacity on Microsoft's Azure cloud, while Microsoft and Nvidia agreed to invest up to $5B and $10B in the company.
Those agreements gave Anthropic three large suppliers instead of one. They also tied it to three companies that sell AI models and assistants of their own to many of the same customers. A cloud customer that doubles as a rival is a different relationship from a tenant who only pays rent.
OpenAI has faced a version of this problem. It started out leaning heavily on Microsoft, then added Oracle and CoreWeave and began building its own Stargate data centers. Anthropic's route to the same independence looks harder, since its main backers are also its main landlords.
The open question is whether Anthropic answers the squeeze by buying more from companies that compete with it, or by committing the kind of capital to its own facilities that an IPO could make possible.