Microsoft unveils Maia 300 AI chip to reduce reliance on Nvidia

15:39

By: Azzat Manal

Microsoft unveils Maia 300 AI chip to reduce reliance on Nvidia

Microsoft is expanding its efforts to develop in-house artificial intelligence hardware as it seeks greater control over computing capacity, lower infrastructure costs and reduce its dependence on Nvidia’s processors.

On Monday, the technology company unveiled Maia 300, the third generation of its custom-designed AI chips. Production is expected to begin in 2027, as Microsoft works to expand computing capacity and improve energy efficiency across its data centres.

The company plans to equip each server rack with 96 Maia 300 chips, double the number supported by the previous generation, according to the announced specifications. Microsoft also says the new system could process up to seven times the volume of tokens used in AI applications, highlighting its focus on handling growing workloads.

Energy efficiency is another central feature of the new hardware. Microsoft expects Maia 300 to deliver a 40% improvement in token processing per megawatt compared with graphics processing units (GPUs). If achieved under comparable operating conditions, this could help reduce the electricity costs associated with running large-scale AI services.

The announcement reflects a broader shift among major technology companies towards designing their own specialised processors. As demand for AI assistants, cloud computing and advanced models increases, hardware performance and access to computing resources have become increasingly important competitive factors.

Nvidia remains a major supplier of AI accelerators, but developing proprietary chips could give Microsoft more flexibility in how it allocates computing workloads across its infrastructure. It could also help the company manage supply constraints and limit its exposure to the cost of purchasing processors from external manufacturers.

The strategy is particularly relevant as data centres face rising electricity requirements and substantial investment needs. Improving the amount of AI work completed per unit of power could help operators control costs, although actual savings will depend on performance in real-world workloads and the wider infrastructure required to support the chips.

Microsoft’s investment in custom silicon does not necessarily mean it will stop using Nvidia hardware. Instead, a broader mix of processors could allow the company to match different chips to specific tasks, balancing performance, energy consumption and operating expenses.

As competition in artificial intelligence intensifies, the ability to secure sufficient computing capacity is becoming a strategic priority for cloud providers. Maia 300 represents another step in Microsoft’s effort to build a more diversified AI infrastructure and strengthen its position in the rapidly expanding market.