Microsoft plans to mass-produce its next-generation AI accelerator, the Maia 300, with an initial or...

Microsoft plans to mass-produce its next-generation AI accelerator, the Maia 300, with an initial order of over 300,000 chips from TSMC, according to The Information. The company aims to publicly unveil the chip as early as September, with deliveries expected by 2027. The long-term goal is to scale production to over one million units, contingent on supply chain and packaging constraints. Maia 300 represents a significant escalation from the current Maia 200, which is produced in tens of thousands of units, and marks a strategic shift toward reducing reliance on Nvidia GPUs for inference workloads.

This move positions Microsoft to become the first major cloud provider with enough custom inference silicon to influence its own customer pricing, offering AI labs a viable alternative to Nvidia beyond the GPU duopoly. The company intends to pitch Maia 300 to external customers, including Anthropic, though Anthropic is reportedly building its own in-house semiconductor team. The 300,000-unit target is still under negotiation and no formal purchase order has been signed, nor have pricing, performance benchmarks, or confirmed external customers been disclosed.

For builders and investors, this signals a broader trend of hyperscalers vertically integrating silicon to control inference costs and margins. If Microsoft achieves its production goals, it could reshape the economics of AI cloud services, pressuring Nvidia's pricing power and prompting other cloud providers to accelerate their own custom chip efforts. However, the significant technical and supply chain hurdles mean the outcome remains uncertain, and the AI chip market will continue to be volatile until these plans solidify.