Microsoft Terminates Oracle Compute Lease; Older-Generation NVIDIA GPUs No Longer Meet Requirements
Microsoft has been reported to be reassessing its third-party AI compute partnership strategy, having decided not to renew the relevant Oracle contracts and begun shutting down some computing resources leased from Oracle, with older-generation NVIDIA GPUs no longer meeting requirements being an important trigger. Microsoft had previously committed to investing more than $80 billion to lease compute from external cloud service providers, involving emerging cloud vendors such as CoreWeave, Nscale, and Nebius. The report also mentioned that Microsoft is not renewing an expansion plan in Texas worth tens of billions of dollars, has withdrawn from a letter of intent for 1.35 gigawatts of capacity in West Virginia, and canceled an approximately $3 billion Loughton agreement with Nscale due to delays in grid interconnection.
Why read this
AI score 62
Details of Microsoft's adjustments to external compute leases show how GPU generational upgrades and power grid interconnection have become hard constraints in cloud compute partnerships.
Topics
Body · AI translation
On November 8, it was reported that Microsoft is reassessing its third-party AI compute partnership strategy: it has committed more than $80 billion to renting external AI compute, but several arrangements are changing, and older-generation Nvidia GPUs no longer meet requirements, becoming an important trigger for Microsoft's adjustment of its Oracle partnership.
Microsoft had previously decided not to renew the relevant contracts and recently began shutting down some computing resources rented from Oracle. This move is also part of Microsoft's reassessment of its multibillion-dollar cloud computing partnerships.
Back then, Microsoft renting servers from its main competitor Oracle was seen as an early signal of a severe shortage of compute resources in the AI industry. As its AI business expanded rapidly, Microsoft also signed large-scale compute leasing agreements with emerging cloud service providers such as CoreWeave, Nscale and Nebius, committing more than $80 billion to rent computing resources from external cloud providers to support its own AI business.
For Microsoft, GPU generational upgrades and power access are becoming two hard constraints in evaluating leasing contracts: the cost-effectiveness of older GPU clusters is declining, while delivery delays from emerging cloud vendors are dragging down the pace of bringing GPUs online.
Some key managers within Microsoft are dissatisfied with the high costs and project delivery delays of emerging cloud service providers. Around the time it decided not to renew the Oracle agreement, Microsoft strengthened oversight of other data center projects, exited some projects deemed higher-risk, and internally discussed whether to continue signing new cooperation agreements with emerging cloud service providers.
Specifically, in Texas, Microsoft decided not to renew an expansion plan worth tens of billions of dollars, but existing contracts are still being performed; in West Virginia, Microsoft withdrew from a letter of intent for 1.35 gigawatts of capacity; in the UK, Microsoft canceled its roughly $3 billion Loughton agreement with Nscale because grid connection delays pushed the expected power supply time into the 2030s.
On Friday local time in Europe, a group of protesters broke into a Microsoft data center construction site in Amsterdam, kicking off a two-day rally. This is part of a growing wave of resistance worldwide to the rapid expansion of computing power required for AI.
Institutions such as Goldman Sachs have warned that in the future Congress will step up investigations into high-tech giants and force them to bear the costs of grid upgrades and renovations. By then, as regulatory scrutiny over AI data centers, model safety, antitrust and other issues intensifies, the capital expenditures and valuations of tech giants will come under pressure.
At the market level, investors recently sold off shares of data center operators such as CoreWeave and Nebius rapidly. Musk also predicted that "more than 200 new cloud platforms will die off and consolidate like small fry."
In response to external concerns, OpenAI Chief Financial Officer Sarah Friar said that although the Democrats' landslide victory in the U.S. midterm elections has raised concerns that data center facilities will face stricter scrutiny, OpenAI will still push ahead with its U.S. data center expansion plan.
AI Trends