Microsoft Orders 300,000 Custom AI Chips for 2027 Delivery

Microsoft is substantially expanding its custom AI chip production, ordering manufacturing capacity from Taiwan Semiconductor Manufacturing Co. for more than 300,000 next-generation Maia accelerators scheduled for delivery in 2027, according to a report from the Information.

The company plans to introduce its Maia 300 chip this fall, representing the third iteration of its AI accelerator line. Microsoft began deploying the Maia 200, its second-generation chip, earlier in 2025.

Why it matters

The massive production commitment signals Microsoft's determination to reduce dependence on Nvidia's expensive and supply-constrained GPUs while competing with cloud rivals that have years-long head starts in custom silicon. With AI workloads consuming enormous computing resources, controlling chip design and supply chains has become a strategic imperative for hyperscale cloud providers seeking to offer customers better economics and performance.

Playing catch-up in custom silicon

Microsoft entered the custom AI chip arena significantly later than its primary cloud competitors. Amazon and Google began developing proprietary accelerators years earlier, and their chips now power substantial AI infrastructure.

Anthropic, a leading AI research company, has committed to using 1 million chips from both Amazon's Trainium line and Google's Tensor Processing Units. This scale underscores how far Microsoft's Maia chips must go to achieve comparable market penetration—the Maia line is generally considered less mature than competing offerings from Amazon and Google.

The economics driving custom chips

All three cloud computing giants share the same fundamental motivation: creating AI accelerators that can be integrated into data centers to deliver cost savings and operational efficiencies to customers. Nvidia's chips remain the gold standard for AI training and inference, but their high prices and limited availability have created intense pressure to develop viable alternatives.

Custom chips allow cloud providers to optimize for their specific workloads, potentially offering better performance-per-dollar for certain AI tasks. They also provide supply chain security and reduce exposure to a single vendor.

A Microsoft manager told the Information that the company's long-term goal is producing gigawatts' worth of Maia chips, though the manager declined to discuss specific production timelines. Microsoft shares rose approximately 1.3% in morning trading following the report.

Production scale and timeline

The 300,000-chip order for 2027 represents a substantial manufacturing commitment. Securing TSMC capacity years in advance reflects both the complexity of advanced chip production and the intense competition for the foundry's limited manufacturing slots.

TSMC produces the most advanced semiconductors globally, and demand for its cutting-edge manufacturing processes far exceeds supply. Major technology companies must book capacity well ahead of their product launches.

The details were first reported by the Information, citing people familiar with the production discussions.