Microsoft could unveil its next-generation Maia artificial intelligence accelerator as early as September, according to a Reuters report Monday citing The Information.
The reported Maia 300 would extend Microsoft’s effort to design more of the hardware beneath Azure’s AI services rather than relying entirely on outside chip suppliers. Reuters reported that Microsoft is discussing production with Taiwan Semiconductor Manufacturing Co. and could seek hundreds of thousands of chips by 2027, though manufacturing capacity and component supply remain potential constraints.
Microsoft has not publicly announced the Maia 300. Its current in-house accelerator, the Maia 200, was introduced in January as an inference-focused chip built on TSMC’s 3-nanometer process. Microsoft said at the time that Maia 200 would support workloads including Microsoft Foundry, Microsoft 365 Copilot and OpenAI models. Microsoft described Maia 200 as part of a broader heterogeneous infrastructure strategy.
The bigger competition is increasingly about the economics of running AI at cloud scale. Alphabet, Amazon and Microsoft are all investing in proprietary accelerators as the cost of advanced AI workloads keeps climbing and demand for Nvidia hardware remains intense.
Why it matters
Custom chips give cloud providers another lever over cost, capacity and performance. If Microsoft can scale Maia beyond internal deployments, the AI infrastructure contest becomes less about buying the best available accelerator and more about controlling the stack from silicon through software.