Microsoft may unveil its next in-house AI chip, the Maia 300, as soon as September, as part of a push to rely less on Nvidia’s processors. Microsoft unveiled its first Maia chip in November 2023 but has been slower than some rivals to scale an in-house alternative to Nvidia. The Information says Microsoft is in talks with TSMC to reserve manufacturing capacity for more than 300,000 Maia 300 chips to be delivered in 2027, with an eventual goal of more than 1 million, though parts supply and negotiations could slow that ramp. Pre-booking production can turn “we have a chip design” into “we can actually ship enough chips to matter.” Google has begun booking revenue from direct sales of its Tensor Processing Units, while Amazon is pushing adoption of its Trainium chips, raising the odds that big cloud providers diversify away from Nvidia-only AI fleets. Microsoft’s 300,000-chip 2027 target is a bet on manufacturing access. If Microsoft can lock in enough TSMC capacity to deliver more than 300,000 Maia 300 chips in 2027 – and scale toward 1 million – it gets a more dependable internal supply of AI accelerators. That reliability can improve visibility into Azure’s AI unit economics, reducing exposure to Nvidia’s pricing and allocation decisions. Even modest substitution gives hyperscalers leverage in supplier negotiations, which could slowly cap how much Nvidia can charge the