Are Microsoft’s AI plans being held back by a shortage of chips?

A Guardian investigation has uncovered a discrepancy between Microsoft's reported AI capabilities and the actual number of advanced AI chips installed in its data centers worldwide. Microsoft aimed to have 1.8 million AI chips in operation by the end of 2024, but recent internal documents reveal that only 2.2 million chips have been installed, less than half of what was anticipated. This indicates that Microsoft's newest data centers may not be fully operational or may lack the necessary chips to keep up with demand.

The shortage of chips, manufactured by Nvidia, contributes to the challenge of keeping pace in the AI arms race. Nvidia's supply chain is closely guarded, and little information is disclosed about its sales or clients. Consequently, it is difficult to accurately assess the growth of AI technologies. Microsoft, one of the leading tech companies, has poured over $280 billion into building AI infrastructure since 2022, but it is challenging to determine the precise number of data centers constructed with this investment.

Microsoft claims to have added 5GW of data center capacity over the past two years, with a total capacity of up to 10GW. However, the sustainability reports suggest Microsoft's AI capacity in 2024 is closer to 1.2GW. This lower figure would imply Microsoft needs approximately 4 million AI chips, assuming it added 5GW of AI data centers in the past two years.

Despite Microsoft's assertions, sources within the company claim that the number of AI chips "barely moved" over the past year. The discrepancy may be partly explained by Microsoft's partnership with OpenAI, which might account for some data center deployments not reflected in the internal documents.