The growing demand for artificial intelligence is fundamentally transforming the data center landscape. For now, Nvidia’s high-performance accelerators continue to dominate the infrastructure behind massive AI workloads. However, that dominance may not last much longer, as Microsoft and other major tech players invest billions in creating their own chips and scaling capacity to match surging computational needs.
At present, Microsoft still depends heavily on Nvidia’s AI chips to power its data centers, but the company is already preparing to rely on its own hardware in the future. Chief Technology Officer Kevin Scott revealed during Italian Tech Week — in a conversation moderated by CNBC hosts — that Microsoft is working on long-term plans aimed at reducing dependency on third-party chipmakers.
Scott explained that Microsoft does not commit to any single AI accelerator but rather chooses the most effective option based on performance and cost efficiency. While Nvidia’s GPU technology has long satisfied Microsoft’s requirements, he stressed that the company remains open to any hardware solution that can deliver the necessary capacity to support growing AI workloads.
Currently, Microsoft operates a diverse mix of accelerators, including chips from Nvidia, AMD, and other partners, alongside its own in-house hardware projects. These include the Arm-based Cobalt CPU and the Maia AI accelerator, both designed specifically for the company’s large-scale data center operations. Scott also confirmed that Microsoft is developing next-generation versions of these chips but did not share technical specifics.
Microsoft already utilizes several custom silicon designs across its data centers and intends to expand these efforts further. According to Scott, the ultimate goal is to create fully integrated systems — encompassing compute, networking, and cooling — to allow complete optimization and control over performance.
The company has become increasingly vocal about its data center strategy. Recently, Microsoft announced plans for what it describes as the most powerful AI data center ever built, expected to outperform today’s leading supercomputers. Part of this effort includes developing an advanced microfluidic cooling technology to improve efficiency and thermal management.
Despite growing concerns over a potential AI-driven market bubble, Scott emphasized that global computing capacity remains insufficient to meet current and future demand. He warned of an impending “massive” capacity shortage, as companies worldwide race to expand their infrastructure to handle the explosive growth of AI workloads sparked by ChatGPT and similar technologies.
