Microsoft is targeting September for the public unveiling of its next-generation Maia 300 artificial intelligence accelerator and is in talks with Taiwan Semiconductor Manufacturing Company (TSMC) to secure production capacity for more than 300,000 units from 2027, according to The Information.
The discussions mark a significant expansion of Microsoft’s in-house AI chip programme as the technology giant seeks to reduce its reliance on Nvidia for the computing hardware powering its artificial intelligence services.
Microsoft is negotiating with TSMC for capacity that would substantially exceed the tens of thousands of Maia 200 chips produced so far. The company is reportedly considering longer-term capacity for more than one million Maia chips, although shortages of components and ongoing negotiations over advanced packaging could limit the scale of the planned production.
The reported September unveiling would represent a major test for Microsoft’s Maia programme, which has faced delays and a limited deployment since the introduction of the Maia 200.
The Maia 200 was delayed after early testing failed to meet Microsoft’s internal performance targets and has subsequently been deployed in only a small number of data centres. Microsoft Chief Executive Satya Nadella told investors during the company’s fourth-quarter fiscal 2026 earnings call on 29 July that Maia 200 delivers 30 per cent better performance per dollar than existing hardware and is being scaled to support OpenAI and Microsoft’s own MAI models.
Microsoft’s broader strategy is to develop proprietary silicon capable of running both its in-house AI systems and OpenAI models at lower cost while reducing its dependence on Nvidia’s processors.
The company is increasing internal use of Maia through Azure AI Foundry and Copilot and is also seeking to attract major external cloud customers. Anthropic is reportedly among the companies Microsoft hopes to secure as a customer for its Maia technology.
However, Anthropic’s own plans could complicate Microsoft’s ambitions. The AI company confirmed earlier this month that it is establishing an internal semiconductor team to develop custom chips for its Claude models, potentially positioning it as both a prospective Maia customer and a future competitor in AI chip development.
Microsoft has the financial capacity to continue investing heavily in its AI infrastructure. Revenue for the fourth quarter of fiscal 2026 reached $90 billion, an 18 per cent increase year-on-year, while Azure and other cloud services revenue rose 43 per cent.
Production remains a key risk to Microsoft’s Maia 300 ambitions. Analysts at J.P. Morgan have warned that projects relying heavily on TSMC’s N3 manufacturing process and CoWoS advanced packaging technology could face supply constraints through 2027.
Microsoft has not publicly confirmed the manufacturing process or packaging configuration planned for Maia 300. The reported September unveiling is also based on information from people familiar with the plans rather than an official Microsoft announcement.
Microsoft shares were indicated about 0.7 per cent higher in pre-market trading on Monday.
The company’s plans will come under further scrutiny as the AI semiconductor market intensifies, particularly ahead of Nvidia’s expected third-quarter fiscal 2027 results on 26 August. Investors are likely to focus on Nvidia’s assessment of competition from custom AI chips developed by major cloud providers.
If Microsoft unveils Maia 300 in September as reported, the launch will provide an early indication of whether the company’s home-grown silicon strategy can overcome the production and deployment challenges that have limited the impact of its first-generation accelerators.
