Microsoft is preparing to expand Maia 300, aiming for more than 300,000 custom AI chips from TSMC

Microsoft is preparing to increase the scale of its custom AI hardware program with Maia 300, according to a report from Yahoo Finance. The story says the company could unveil the next Maia model as soon as September, and that Microsoft is reportedly working to secure production capacity from Taiwan Semiconductor Manufacturing Co. for more than 300,000 chips.

Maia is Microsoft’s internal line of custom silicon designed to run parts of its AI workloads, with the intent of improving performance and controlling costs compared with relying solely on third-party accelerator supplies. Microsoft has been positioning its chip efforts as part of a broader push to build and operate large-scale AI infrastructure through a combination of data centers, cloud software, and specialized hardware.

The new report’s most concrete figure is the manufacturing request, which points to a ramp that goes beyond a single test-and-learn cycle. Seeking production for more than 300,000 chips suggests Microsoft wants Maia 300 to play a larger role in its AI computing mix rather than remaining a limited deployment.

The report also frames timing as a key element. It says Microsoft could reveal Maia 300 by September, which would place the rollout of a new Maia generation relatively near-term. For companies building AI services, calendar milestones like these matter because data center deployments and model training schedules often require months of lead time for hardware procurement and installation.

While the Yahoo report discusses the prospective manufacturing scale and the possible unveiling window, it does not provide additional technical details in the information available for this review. For example, it does not specify Maia 300’s architecture, performance targets, power characteristics, or the exact mix of customer workloads or internal training tasks it is intended to support first.

Microsoft also did not appear in the available material with a direct quote or formal disclosure about Maia 300’s production commitments or manufacturing orders. In the absence of a primary company statement or a regulatory filing, the reported manufacturing and timing points should be treated as expectations attributed to the reporting rather than confirmed program numbers.

The broader context is that large cloud and AI operators are trying to reduce friction in the supply chain for AI accelerators. Custom chips can help manage availability and unit economics, but they require careful coordination with foundries and packaging partners. A reported step-up in the number of chips would be consistent with a company that believes its custom hardware is moving from pilot usage toward larger-scale deployment.

What to watch next is whether Microsoft confirms Maia 300’s timing, scope, and production quantities in an official channel, and whether additional details emerge about how it is allocating Maia capacity across training, inference (running trained models), or both. Investors and customers will also likely monitor whether Microsoft’s custom hardware roadmap changes in response to market demand for AI compute and the availability of third-party accelerator alternatives.