Microsoft has reportedly pulled out of lease agreements for building data centre capacity in the United States of America, fuelling speculation that the tech giant is rethinking its expectations of AI demand in a post-DeepSeek era. Analysts at investment banking company TD Cowen found that Microsoft has cancelled leases with private contractors for establishing “a couple of hundred megawatts” of data centre capacity in the US, according to a report by Bloomberg. The TD Cowen analysts reportedly made inquiries with supply chain providers to uncover signs that Microsoft let agreements for data centres with more than a gigawatt of capacity expire. It has also backed away from multiple deals for building out data centres with 100 megawatts capacity each, as per the report. Why has Microsoft terminated data centre leases? Microsoft is one of the largest owners and operators of data centres in the world, and has spent billions of dollars in the past to secure AI computing capacity. While the reason for Microsoft’s purported move is not clear, TD Cowen said that the company cited "used facility" and "power delays" as reasons for terminating the data centre leases. The analyst report further suggested that Microsoft may be reallocating some of its AI data centre investments from overseas countries to the US. “While we have yet to get the level of colour via our channel checks that we would like into why this is occurring, our initial reaction is that this is tied to Microsoft potentially being in an oversupply position,” TD Cowen analysts Michael Elias, Cooper Belanger, and Gregory Williams were quoted as saying by Bloomberg. In a second report on Monday, February 24, TD Cowen said OpenAI has been relocating workloads from Microsoft servers to Oracle servers as part of a new partnership with the Larry Ellison-led firm. Microsoft counts itself as OpenAI’s biggest backer and had a special arrangement with the ChatGPT-maker to exclusively host the latter’s AI models on its servers. The arrangement was altered earlier this year to allow OpenAI to use cloud-computing services from rival providers. However, Microsoft still has a right of first refusal when the AI startup seeks computing power from other providers. In response to the TD Cowen report, Microsoft reiterated its $80 billion commitment on AI data centres this fiscal year. “While we may strategically pace or adjust our infrastructure in some areas, we will continue to grow strongly in all regions. Our plans to spend over $80 billion on infrastructure this FY remains on track as we continue to grow at a record pace to meet customer demand,” a Microsoft spokesperson was quoted as saying. Has DeepSeek hurt AI chip demand? The emergence of DeepSeek was a major inflection point for AI development. The Chinese AI startup claims that its R1 model (rivalling OpenAI’s o1) took fewer resources and less time to build than what was previously believed. This has sparked fresh skepticism of the massive AI spending plans by tech giants such as Google, Microsoft, and OpenAI. In the wake of the DeepSeek-driven market turmoil, Microsoft CEO Satya Nadella had argued that the Chinese AI startup’s impact could, counterintuitively, increase demand for advanced GPUs. Nvidia founder and CEO Jensen Huang has also asserted that DeepSeek’s technological advancements will not negatively impact the chip giant’s business. “I think the market responded to R1, as in, ‘Oh my gosh. AI is finished’. You know, it dropped out of the sky. We don’t need to do any computing anymore. It’s exactly the opposite. It’s [the] complete opposite,” Huang said in an interview with Alex Bouzari, CEO of DataDirect Networks, on February 20. “It’s making everybody take notice that, okay, there are opportunities to have the models be far more efficient than what we thought was possible. And so it’s expanding, and it’s accelerating the adoption of AI,” Huang said, adding that AI reasoning models are still “fairly compute-intensive.” Following the news of DeepSeek, Nvidia’s share price plummeted 16.9 per cent, the largest single-day decline for a public company, and nearly $600 billion was wiped off from the American chipmaker’s market value. What does Microsoft’s purported decision mean? Big tech companies such as Microsoft, Meta, Amazon, and Google projected spending a total of $320 billion in FY2025 to build data centres and purchase clusters of advanced graphics processing units (GPUs) to power their AI models. However, Microsoft’s purported lease withdrawals have re-ignited concerns about whether AI demand projections are overstated. While tech giants have doubled down on their AI spending commitments, critics have cautioned that the real-world use cases of AI and paths for monetisation remain unclear. Nadella himself has said that "at some point, the supply and demand [for AI] have to map." Appearing in a podcast interview on YouTube last week, he said, “You can go off the rails completely when you are hyping yourself with the supply-side, versus really understanding how to translate that into real value to customers.” “Us self-claiming some AGI milestone, that’s just nonsensical benchmark hacking to me. The real benchmark is the world growing at 10 per cent," Nadella added. However, Nadella also maintained his position that the computing need of the world for developing and running AI models will grow exponentially. “Anytime there’s a breakthrough like even what DeepSeek did with the efficient frontier of performance per token changes, the curve gets bent and the frontier moves. That just brings more demand,” he said. In January this year, US President Donald Trump unveiled the Stargate Project with Microsoft, OpenAI, Nvidia, Arm, Oracle, and Softbank as corporate partners in the joint venture that aims to invest $500 billion to build out AI infrastructure such as data centres, energy plants, power lines, and more in the US over the next four years. On Tuesday, February 25, The Information reported that Meta is in talks to construct a new data centre campus for its AI projects, with potential costs exceeding $200 billion.
