
律动BlockBeats|5月 22, 2026 05:25
[Microsoft's Computing Power Shifts Toward Anthropic, Both Parties Negotiating Maia Custom Chip Rental Agreement]
According to monitoring by Beating, Anthropic is in early discussions with Microsoft regarding the rental of Microsoft's custom AI chip, Maia, to secure additional computing power support. If an agreement is reached, the collaboration between the two parties will deepen further, and Anthropic will become a major customer of Microsoft's Maia 200 chips. Microsoft began deploying Maia 200 chips in Azure cloud data centers in January 2026, primarily to reduce the operating costs of Copilot agents. Renting custom chips aligns with Anthropic's long-term multi-chip hardware strategy. Unlike OpenAI, which relies solely on NVIDIA chips, Anthropic has long rented Google's TPUs and Amazon's custom chips and is currently in talks with a UK-based chip startup for potential collaboration.
On the financial front, although Anthropic is expected to achieve nearly $11 billion in revenue and $560 million in operating profit this quarter, it has committed to at least $330 billion in cloud spending with the three major U.S. cloud service providers, including $30 billion to Azure. Utilizing custom chips from cloud providers not only offers Anthropic the advantage of tailored hardware but also alleviates the burden of high computing costs through cost subsidies from cloud providers.
The relationship between Microsoft and Anthropic has been heating up rapidly in recent months. At the end of last year, Microsoft invested as much as $5 billion in Anthropic and has since become a key customer, planning to purchase at least $500 million worth of Claude model usage rights to power its Copilot agents. To meet Anthropic's surging demand for computing power, Microsoft has been steadily increasing its resource allocation to Anthropic since November 2025, not only reallocating more existing NVIDIA servers to Anthropic but also building entirely new server clusters for them. In contrast, the computing power needs of some smaller cloud customers have been selectively delayed. [Original Link]
Timeline