Microsoft announced the Maia 300 chip and plans for deployment from 2027
Microsoft announced Maia 300, the third generation of its AI chip, with production deployment planned from 2027. It plans to expand the Maia family to several gigawatts and use it as the primary system for inference. According to the company, Maia 300 is expected to offer 40 % more tokens per megawatt than leading GPU-based systems.
Microsoft announced Maia 300, the third generation of its own AI chip. Production deployment is planned to begin in 2027. Starting in the same year, the company plans to expand infrastructure based on the Maia family to several gigawatts and make it the primary system for inference in products and services from Microsoft.
According to Microsoft, Maia 300 is expected to deliver 40 % more tokens per megawatt than leading GPU-based systems. The company says the full configuration with 96 accelerators is expected to achieve up to seven times the token throughput of a Maia 200 system. These are expected specifications announced by the manufacturer.
The chiplet architecture uses the N2P manufacturing process from TSMC and separates compute components, the memory controller and I/O functions. According to Microsoft, this arrangement is intended to increase the chip area dedicated to computation. The company says the system is designed for long-context reasoning, latency-sensitive conversational services and long-running, multi-step agent tasks.
Why it matters
If the company's expectations are borne out, Maia 300 will enable more tokens to be processed at the same power consumption. The planned deployment at a scale of several gigawatts thus links energy efficiency to the capacity of future AI services. However, the comparisons cited measure different properties: energy efficiency against GPU-based systems and the throughput of the full configuration against a Maia 200 system.
What was added since the original report
Verified updates
-
A rack with 96 Maia 300 chips is expected to deliver up to seven times as many tokens as a Maia 200 rack.; Maia 300 is intended to target long inputs, fast chats and agent tasks.
- A rack with 96 Maia 300 chips is expected to deliver up to seven times as many tokens as a Maia 200 rack.
- Maia 300 is intended to target long inputs, fast chats and agent tasks.
Relevant practical impact
What this means
For a business
For Microsoft, this is a plan to change its cloud infrastructure: its own chips are intended to take over the bulk of inference workloads, with the aim of reducing dependence on GPUs from Nvidia and AMD.
Strategy More business impacts →Check the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.