Anthropic explores Microsoft AI chips for Claude

Anthropic is in early talks to rent Microsoft’s Maia AI chips to add computing power for Claude amid rising demand.

The talks may not lead to a deal.

Anthropic agreed in 2025 to buy US$30 billion of Azure computing capacity.

Microsoft has offered Anthropic models through its cloud service.

Anthropic has also signed large chip deals with Google and Amazon.

Microsoft is building in-house processors to cut its reliance on Nvidia.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Microsoft’s Maia 100 beats Nvidia’s older A100 graphics processing unit (GPU), a chip used for AI computing, on some inference tasks. It cuts latency by 12% and uses 18% less power 1.
Maia does not replace Nvidia’s top chips for every job. In large model training, it delivers 6% less throughput than an H100, while reaching similar convergence times 1.
The chip has 64 gigabytes of High Bandwidth Memory (HBM), a type of fast memory used in AI chips. That is less than recent Nvidia or Advanced Micro Devices (AMD) parts, so Maia fits smaller or more targeted AI models 2.
Its edge comes from tight ties to the Azure software stack. That can lift performance for Azure cloud customers and make deployment simpler 1.

Anthropic has looked at renting Microsoft’s Maia chips even though Amazon is still its main cloud provider and training partner. That suggests AI model builders want more hardware and cloud options for extra capacity and less reliance on one vendor 3.
The move fits a wider “circular spending” pattern. Large tech companies put billions into AI labs, then those labs sign huge computing contracts with the same backers 4.
For Microsoft, selling Maia beside Nvidia and Advanced Micro Devices (AMD) products gives it more pricing leverage. It also makes Azure a more flexible cloud platform for customers 2.