Source: AFP

Microsoft announced on Monday that it intends to give priority to its own AI chips for running its AI products and services, a move that mirrors what its cloud rivals Amazon and Google have already been doing for years. The company unveiled the Maia 300, its third generation of dedicated AI processors, and said the chip is expected to be operational in 2027. Microsoft says the Maia 300 can allow AI to generate seven times the content of its predecessor, and that at equivalent electricity consumption it delivers around 2.5 times the performance per watt of the Maia 200.

Custom silicon is now a cloud strategy, not a side project

Microsoft is now proposing a fully integrated architecture around Maia, from the chip through servers and internal networking to the data centre. The chips are focused on inference, the phase of AI where trained models actually serve users, which is by far the fastest growing and most cost sensitive part of AI workloads. Using its own silicon allows Microsoft to lower costs and tailor processors to its own workloads rather than paying a premium for general purpose GPUs. For now Microsoft does not sell the chips separately, they are reserved for its own systems, while Google has started offering its TPUs to external customers and AWS is reported to be in discussions with potential partners.

Why this matters for teams building on Azure

For engineering teams, this is a signal about where cloud pricing and capacity are heading. When a hyperscaler optimises its own infrastructure for inference, the cost benefits tend to appear in the services customers consume, from AI APIs to platform features, and capacity for inference heavy workloads becomes more predictable. Teams designing architectures today should avoid hard coupling to a single accelerator or provider, keep inference layers abstracted so workloads can follow the best price to performance, and treat cloud cost review as a standing part of the DevOps cycle rather than a quarterly afterthought.

If you want help designing cloud architecture that stays portable across providers, optimising AI workload costs or reviewing your current Azure setup, contact Excello Digital at https://excello.digital/contact/. We help European companies build infrastructure that takes advantage of these shifts instead of paying for them.