i
News
News · 2026-10-12

Microsoft targets several gigawatts of Maia chips by 2027

@neuronium_ai @neuronium_ai

Microsoft says it plans to scale its Maia AI-chip capacity to several gigawatts by 2027, when it expects to ship Maia 300. The company is pitching the next generation for long inputs, fast conversations and AI-agent workloads, and claims higher token output per unit of power than leading GPU systems. Those comparisons make the announcement more than a chip roadmap: they are Microsoft’s case for building its own infrastructure at scale, though the figures come from the company and the technical report has not been included here.

Cover: Microsoft targets several gigawatts of Maia chips by 2027
Each Maia 300 chip packs eight HBM4 memory stacks totaling 288 GB with 18 TB/s of bandwidth. It's rated at 950 watts. | Image: Microsoft

Each Maia 300 chip packs eight HBM4 memory stacks totaling 288 GB with 18 TB/s of bandwidth. It's rated at 950 watts. | Image: Microsoft

Source: the-decoder.com

What Microsoft says Maia 300 will deliver

Microsoft says Maia 300 will produce 40% more tokens per megawatt than leading GPU systems. A rack containing 96 Maia 300 chips, the company says, should deliver up to seven times as many tokens as a Maia 200 rack.

The details are in a technical report. Microsoft says it will present more material at Ignite 2026 in November.

The claim that needs context

Tokens per megawatt is a useful measure of throughput against power use, but it does not tell us how Maia 300 compares on price, availability or performance across the workloads Microsoft names. I think the headline figures are compelling, but without those details they are not enough to judge whether the chips will be a practical alternative to leading GPU systems.

The other notable gap is what “several gigawatts” means in deployment terms. The announcement gives a target for Microsoft’s own Maia capacity by 2027, not a breakdown of how much will be Maia 300 or where that capacity will be used. Until the technical report and further details arrive, the scale is clearer than the economics.

Source: the-decoder.com

Daily AI news

Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.

Only what matters — every day

Follow on X