Microsoft is preparing to expand its use of in-house artificial intelligence hardware with the introduction of Maia 300, a new chip designed to improve computing efficiency and reduce its reliance on external chip suppliers.
The third-generation processor is expected to begin handling production AI workloads in 2027, supporting Microsoft's efforts to power its cloud services and AI products with its own infrastructure.
The announcement marks another step in the company's strategy to develop specialised hardware capable of handling increasingly complex AI tasks while controlling the cost of operating large-scale systems.
New chip promises significant performance gains
Microsoft says Maia 300 is designed to deliver greater efficiency when running AI models, particularly those that require advanced reasoning, rapid responses and multiple connected tasks.
Scott Guthrie, executive vice president of Microsoft's cloud group, said the company expects the processor to deliver 40 per cent more tokens per megawatt than leading GPU-based systems.
Tokens are the units of text or other data that AI models process and generate. Higher token output relative to electricity consumption can help cloud operators improve computing efficiency.
Microsoft also projects that a complete configuration containing 96 accelerators could deliver up to seven times the rack-scale token throughput of Maia 200.
These figures are Microsoft's stated performance expectations, rather than independently verified comparisons.
Microsoft seeks greater control over AI infrastructure
The development reflects Microsoft's ambition to reduce its dependence on third-party processors, particularly graphics processing units (GPUs) supplied by Nvidia and AMD.
Building its own AI chips gives the company greater control over the hardware supporting its cloud services. It could also help Microsoft optimise computing resources for its specific workloads.
Guthrie said Microsoft plans to expand its Azure Maia infrastructure to deployments measured in multiple gigawatts from 2027.
The company intends for the Maia family to become its primary system for running AI inference across Microsoft products and services.
AI inference refers to the process of using a trained model to generate answers, predictions or other outputs in response to user requests.
Maia 300 builds on Microsoft's existing AI chip programme
Maia 300 follows Maia 200, which Microsoft announced in January 2026.
According to the company, the earlier processor is already supporting production workloads across several of its services, including:
- Microsoft Copilot, its AI assistant.
- GitHub Copilot, its AI-powered coding assistant.
- Microsoft Foundry, its platform for developing and deploying AI applications.
- Microsoft's internally developed AI models.
The experience gained from deploying Maia 200 provides a foundation for the next generation of the company's custom silicon.
Microsoft says Maia 300 has been designed for emerging AI workloads that place greater demands on computing infrastructure.
These include long-context reasoning, conversational services that require quick responses and agentic systems capable of carrying out multiple connected steps.
Chiplet architecture focuses on efficiency
A key element of Maia 300 is its use of a chiplet-based architecture, which Microsoft says will help deliver important efficiency improvements.
Chiplet designs use multiple smaller semiconductor components within a larger processor package rather than relying exclusively on one large, integrated chip.
This approach can offer manufacturers greater flexibility when designing complex processors, although the actual benefits depend on how the components are built and integrated.
For Microsoft, efficiency is particularly important as demand for AI services increases the amount of computing power required in data centres.
Improving the volume of AI work completed with a given amount of electricity could help the company manage infrastructure costs as it expands its services.
What the new chip means for the AI industry
Microsoft's investment in custom processors highlights a broader shift among major technology companies towards developing their own AI hardware.
As businesses integrate AI into search, productivity software, coding tools and cloud services, the cost of running these systems has become an increasingly important consideration.
Custom chips offer companies an opportunity to tailor hardware to their workloads rather than relying entirely on commercially available processors.
However, Microsoft's announcement does not mean that Nvidia and AMD hardware will immediately disappear from its infrastructure. The company has outlined a plan to make Maia its primary AI inference system, but it has not announced a complete end to using third-party chips.
The next major milestone will be Maia 300's planned production rollout in 2027. Its real-world performance, deployment scale, and operating costs will help determine how significantly it changes Microsoft's AI infrastructure strategy.
Source: Microsoft