The hunt for cheaper AI compute is pushing Microsoft to ramp its in-house Maia accelerators, with a plan to reveal the Maia 300 as soon as next month.
The Information reported the roadmap, citing people familiar with Microsoft’s plans, and the company has not commented.
Behind the scenes, Microsoft is in talks with TSMC about manufacturing capacity for the accelerator, according to the report.
One volume target under discussion is more than 300,000 chips, which Microsoft wants delivered during 2027.
Its longer-term ambition stretches past 1 million units, though supply talks could slow the ramp.
Microsoft shipped the second-generation Maia 200 in January, years after introducing the first Maia part in late 2023, and still trails Alphabet and Amazon in scaling custom silicon.
It is also courting big cloud customers, Anthropic among them, to run workloads on Maia.
The timing matters because Nvidia’s AI GPUs stay scarce and expensive, and DigiTimes reports Nvidia still controls about 60% of the advanced packaging queue.
A successful ramp would hand Microsoft a genuine alternative to Nvidia inside its own data centers and fresh leverage over the wider AI buildout.
