Anthropic is in negotiations with Microsoft to lease servers powered by Microsoft's custom Maia AI chip. A successful agreement would mark a significant strategic milestone for Microsoft's in-house chip efforts and could reshape the competitive landscape of the cloud computing market.
According to recent reports, Anthropic is seeking to lease Maia chip servers to secure additional computing power to meet the growing demands of its Claude model. Should the talks succeed, Microsoft would secure a major external client, providing a substantial boost to its long-term efforts to catch up with Google and Amazon in the custom chip arena.
For the broader market, this development signals a deepening cloud partnership between Microsoft and Anthropic. It also reflects a broader industry trend where leading AI players are accelerating efforts to reduce reliance on NVIDIA's hardware.
Microsoft's Maia Chip: A Strategic Move to Catch Up The strategic rationale behind Microsoft's development of the Maia chip mirrors that of Google's TPU and Amazon's Trainium: to create alternatives to NVIDIA's hardware, thereby reducing cloud service hardware costs and protecting profit margins. Cloud executives widely view in-house chip development as essential for maintaining profitability amidst high NVIDIA hardware costs.
However, Microsoft's path has not been smooth. The Maia chip project faced delays last year and remains in a relatively earlier stage compared to offerings from Google and Amazon. In January, Microsoft announced that the latest version, Maia 200, has begun deployment in Azure data centers. Azure head Scott Guthrie stated that Maia 200 is being used to lower the operating costs of the Copilot AI tool, which is powered by models from OpenAI and Anthropic.
In terms of positioning, the Maia chip is designed to run existing models faster than NVIDIA chips but is not intended for training new models from scratch. This focus makes it more suitable for large-scale inference deployment rather than cutting-edge model development.
Anthropic's Multi-Chip Strategy: Avoiding Vendor Lock-In Anthropic has long pursued a differentiated chip procurement strategy. Unlike the heavy reliance on NVIDIA hardware seen at OpenAI and xAI, Anthropic has built a diversified computing infrastructure incorporating chips from Amazon, Google, and NVIDIA. Reports also indicate it is in talks with a UK-based chip startup.
The current negotiations with Microsoft represent a natural extension of this multi-vendor strategy. A deal would provide Anthropic with another computing option for running its Claude models. Furthermore, it could allow Anthropic to influence the design direction of future Maia chip iterations to better suit its specific needs.
Financially, Anthropic has strong incentives to expand its cloud expenditure. Reports suggest its quarterly revenue could approach $110 billion this quarter, more than doubling from the previous quarter, with an expected quarterly profit of around $5.6 billion. The company has already committed to spending a combined minimum of $330 billion with the three major U.S. cloud providers. Analysts note that if cloud providers offer subsidies for using their custom chips, adopting such chips could theoretically provide Anthropic with better cost-performance.
Deepening Ties Between Microsoft and Anthropic The partnership between Microsoft and Anthropic has intensified over the past year, a period that has coincided with OpenAI moving closer to Microsoft's competitors.
Commercially, Microsoft is a significant customer for Anthropic, with plans to purchase at least $5 billion worth of access to the Claude model to power its Copilot products. Earlier this year, Microsoft also began reselling Anthropic's models on its Azure platform alongside OpenAI's models. Late last year, Microsoft announced an investment of up to $50 billion in Anthropic, which in turn committed to $300 billion in cloud spending on Azure.
Regarding computing resource allocation, usage of Azure by Anthropic has been steadily increasing since last November, according to a source with direct knowledge. Microsoft has recently allocated additional existing NVIDIA server capacity to Anthropic and has begun building new, dedicated server clusters for the company, reportedly at the expense of some resources for smaller customers.
The current chip negotiations are seen as a signal that the existing cloud cooperation agreement between the two companies could be expanded in the future. However, the scale of Microsoft's cloud agreement with OpenAI still significantly exceeds its current arrangement with Anthropic.
Multi-Chip Strategy Gains Traction Anthropic's multi-vendor approach is reportedly inspiring imitation. Meta and OpenAI are both attempting to replicate this strategy, having secured agreements with multiple chip suppliers. Concurrently, both companies are also advancing their own in-house chip development plans. Microsoft has reportedly reserved data center space for an upcoming custom chip from OpenAI, though no formal procurement agreement has been reached.
These developments indicate that the competitive landscape of the AI computing market is evolving from NVIDIA's dominance towards a more diversified supply chain. For cloud providers, the ability to attract external clients for their custom chips will be a key metric for measuring the success of their chip strategies. For AI developers, building a diversified portfolio of computing sources is both a practical choice for cost efficiency and a strategic consideration for mitigating supply chain risks.
Comments