According to information from investment bank TD Cowen, Microsoft has terminated several data center leases and scaled back international investments. How come?
At least two locations in the U.S. are known to have had their leases terminated by Microsoft. According to TD Cowen, this represents “a couple hundreds of megawatts.” The information comes from various suppliers and third parties, which the investment bank uses to measure a company’s financial health and transactions.
Signs of scaling back
TD Cowen indicates that Meta showed similar signs when it sought to cut costs. For Microsoft, this reporting contradicts the company’s generous commitments. Earlier this year, Chair and VP Brad Smith told Microsoft that around $80 billion would have been spent on AI infrastructure between mid-2024 and mid-2025 (Microsoft’s financial year). It may be that most of this has already been spent or committed, and that the lease terminations merely represent a realignment.
Nonetheless, one AI major has already been moving away from Microsoft: OpenAI. The company behind ChatGPT is participating in the ambitious Stargate project, which will consist of a $500 billion investment into an AI infrastructure buildout in America. Oracle and SoftBank are also taking part; the latter would also like to put a large sum of money into OpenAI.
This is where the roads separate
An earlier report from The Information already suggested that OpenAI is exiting Microsoft’s Azure cloud. In exchange, SoftBank’s infrastructure will become the new home for OpenAI’s demanding AI workloads. It would mark a significant turnaround from early 2023, when Microsoft had decided to invest $10 billion in the AI player. Since then, the relationship between the two parties has been severely tested. When Sam Altman was briefly ousted as CEO, it seemed for a while that the OpenAI exodus would fuel an AI team within Microsoft. Nothing proved less true, and a Microsoft observer position on the board of directors in mid-2024 disappeared after only six months.
Whether or not the leases are related to OpenAI’s exit is unclear. What is known is that the company’s training and inferencing requirements are enormous. Almost all the money coming in to the AI company goes into computing power and R&D (which also costs computing power). Even the prohibitively expensive ChatGPT Pro represent a loss for OpenAI due to the astronomical cost of running the models, CEO Sam Altman revealed. Without outside backers, OpenAI wouldn’t have made it to 2025 either; SoftBank is currently simply the latest investor keeping it alive.
However, for Microsoft, there is reason enough to invest heavily in AI infrastructure. It recently integrated Copilot functionality into its fixed 365 offering, which will theoretically see millions of users using AI. No doubt, this may require an expansion of Azure capacity. However, whether it should cost so much and whether Microsoft should build it out so quickly is more difficult to determine. Still, the market is feeling the after-effects somewhat over doubts that struck when Chinese DeepSeek arrived on the scene. The thinking behind these doubts is that less compute may be needed than previously thought, as LLMs of a smaller size than the highest-end models still show stunning performance. In response, however, Microsoft CEO Satya Nadella threw up Jevons’ paradox, which states that consumption only increases when new technology is accessible with fewer resources than before.
Also read: Microsoft shadows Google: 365 now includes AI as standard