Microsoft still relies heavily on Nvidia, but its Maia chips show how major cloud companies are beginning to divert inference workloads to home-grown silicon.
Microsoft’s growing push into custom AI silicon highlights a shift that could matter for Nvidia’s long-term sales base. The concern is not that Microsoft will stop buying Nvidia chips altogether. It is that one of Nvidia’s largest customers is increasingly building alternatives for the parts of AI computing it can control, especially inference, the stage where models generate answers and run services at scale. That makes Microsoft both a major buyer and a potential structural source of reduced dependence for Nvidia, according to the company’s own recent disclosures and comments reported by industry outlets.
The scale of that relationship is significant. Nvidia’s revenue remains heavily concentrated among a handful of cloud and platform companies, including Microsoft, Amazon, Super Micro Computer, Alphabet and Meta Platforms, which together account for more than half of sales, according to the figures cited in the lead article. That concentration has helped drive Nvidia’s rise, but it also increases the risk that large buyers will seek their own chips once the market matures enough to support alternative architectures. Microsoft’s Maia programme shows exactly that pattern: a customer that still needs Nvidia, yet is no longer content to rely on it exclusively.
According to Microsoft’s announcement, Maia 300 is designed for inference workloads rather than model training, with an “inference-first” architecture aimed at long-context reasoning, conversational systems and multi-step agentic tasks. Microsoft said the chip will support larger deployments from 2027 and will be rolled out across Azure infrastructure, with the company claiming that a 96-accelerator rack can deliver up to seven times the throughput of Maia 200 and about 40 per cent more tokens per megawatt than leading GPU-based systems. The company also said Maia 200 is already running production workloads, including Microsoft Copilot, with GitHub Copilot and Microsoft Foundry due to follow.
The strategic point, however, is broader than one product launch. Axios reported that the move reflects a wider industry change as large technology groups try to reduce dependence on Nvidia by designing their own silicon. Microsoft’s internal chip effort sits alongside similar work at Google, Amazon and OpenAI, each of which wants more control over cost, power efficiency and supply. Even so, Nvidia remains central to the AI build-out, particularly for training frontier models and for customers that have not yet developed their own hardware. The risk for Nvidia is therefore not an abrupt collapse in demand, but a gradual erosion of its hold on the biggest buyers as they redirect more of their spending to home-grown systems.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
