The company plans to unveil Maia 300 this fall and has been discussing manufacturing capacity with TSMC for 2027 delivery. A Microsoft manager said the ultimate goal is to produce "gigawatts' worth" of Maia chips.
Credit: Microsoft
Microsoft in talks with TSMC for 300,000+ Maia 300 chips for 2027. Maia 300 unveil this fall. Follows Maia 200 launch in January. Ultimate goal: gigawatts of capacity. Anthropic alone uses 1M Amazon and Google chips. Microsoft’s custom silicon is behind rivals.
Microsoft is planning to “significantly” increase production of its next-generation AI chips, The Information reported. The company has been in discussions with TSMC to secure manufacturing capacity for more than 300,000 Maia 300 chips for delivery in 2027. Microsoft plans to unveil Maia 300 this fall, and a company manager said the ultimate goal is to produce “gigawatts’ worth” of Maia chips. The scale of that ambition puts Microsoft on a path to build its own AI compute layer rather than buying everything from Nvidia.
The urgency is real. Anthropic is building its own chip team and has deals to use one million Amazon Trainium chips and Google TPUs. Microsoft’s custom chip programme started years after Amazon and Google began designing their own, and the Maia line is widely seen as less mature than its rivals’ hardware. Maia 200, the second-generation accelerator, only started rolling out in January. The 300,000-unit order for Maia 300 is Microsoft’s attempt to close that gap before its cloud customers start choosing Amazon or Google specifically because of their custom silicon.
All three cloud giants are chasing the same prize: AI accelerators that cost less and run more efficiently than Nvidia’s GPUs. Nvidia’s chips remain the industry standard, but their high prices and chronic short supply have turned custom silicon from a nice-to-have into a strategic necessity. Microsoft recently told employees to stop “tokenmaxxing“ and set division-level AI budgets, a sign that compute costs are becoming a constraint even at the company spending $35.8 billion per quarter on infrastructure.
The challenge is execution. TSMC’s advanced packaging capacity is tight through 2027, and Microsoft needs to persuade enterprise customers that Maia can match or beat Nvidia on the workloads that matter. The longer-term ambition is capacity for more than one million units, but component supplies could constrain that target. For now, 300,000 chips is a bet that Microsoft’s custom silicon will be ready for production scale by 2027. If it works, Azure gets a cost advantage over Nvidia-dependent competitors. If it does not, Microsoft will have spent billions on a chip programme that is still a generation behind.
Get the most important tech news in your inbox each week.
