Microsoft plans to deploy its Maia 300 custom AI chip at multi-gigawatt scale from 2027, positioning it as the primary inference system for its products and services. The announcement signals a major expansion of Microsoft’s in-house AI compute, with potential consequences for how its services are built and supplied.
Microsoft AI Watch analysis
What happened
Microsoft says Maia 300 is its third-generation custom AI chip. The company describes a chiplet design using TSMC’s 2-nanometre process for computing and 3-nanometre technology for data movement, intended for long-context and multi-step AI tasks.
Microsoft claims Maia 300 delivers about 40% more computing efficiency per megawatt than leading GPU-based systems. It also says a 96-chip rack provides seven times the token throughput of Maia 200. Those are the company’s performance claims, not independently established comparisons. Microsoft says Maia 200 is already in production for workloads including Copilot, and it plans to deploy more than a million Maia 300 units starting next year.
Why it matters
A multi-gigawatt rollout would make Maia more than a bespoke chip project: it would put Microsoft’s own silicon at the centre of a large share of its AI inference. That could give the company more control over capacity and efficiency as demand grows, while changing the balance between custom chips and the GPUs it buys from suppliers.
The scale is striking, but plans are not deployed hardware. The useful test will be whether Maia 300 arrives on schedule and whether its claimed gains show up across real workloads, not just in a rack-level comparison.
Our read
This is a substantial infrastructure bet, not a chip unveiling for the spec-sheet scrapbook. If Microsoft can deliver the planned volume and performance, Maia could reshape how it serves AI across its own products. For now, the numbers belong to Microsoft’s announcement; the deployment is the part to watch.
What to watch
- Whether Maia 300 deployments begin on the stated schedule in 2027.
- How much of Microsoft’s inference workload moves to Maia.
- Whether independent results support the company’s efficiency and throughput claims.
Discussion spark: If Microsoft’s Maia chips deliver the promised efficiency at scale, should it prioritise its own silicon over GPUs from outside suppliers?
Sources and evidence
- Microsoft to expand Azure Maia system to multi-gigawatt deployments (12 October 2026, 13:01 UTC)
not affiliated with or endorsed by Microsoft