OpenAI reshapes its cloud computing supply landscape: Capital expenditures on Oracle servers will exceed investments in Microsoft in the coming years

OpenAI is reshaping its cloud computing supply landscape, with the world's highest-valued startup planning to spend more on servers from competitors like Oracle than on its long-time partner Microsoft in the coming years, marking a significant shift in one of the most important partnerships in the artificial intelligence industry.

According to The Information, citing an OpenAI employee familiar with the situation, OpenAI has prepared a budget of up to $450 billion for server spending through 2030. Of this, the company's planned spending on leasing servers from competitors like Oracle from now until the end of 2030 will exceed its investment in Microsoft.

This shift highlights that OpenAI's rapidly growing demand for computing power has surpassed the capacity of Microsoft as a single supplier. Despite Microsoft's priority supply rights due to a $13 billion investment, OpenAI CEO Sam Altman's dissatisfaction with the pace of Microsoft's data center construction prompted both parties to reach an agreement in the summer of 2024, allowing OpenAI to seek other cloud service providers.

For investors, this change reshapes the cloud computing market landscape. Oracle's stock surged 36% after announcing in early September that its cloud business would generate $400 billion in revenue over the next four years, while some Microsoft shareholders expressed a preference for OpenAI to continue exclusively using Azure services. Nevertheless, Microsoft will still receive significant revenue from OpenAI, which plans to invest about $135 billion in Microsoft servers by 2030.

Changes in Microsoft and OpenAI's Partnership

The exclusive agreement between Microsoft and OpenAI faced a critical turning point in 2024. That summer, Altman sought permission from Microsoft to negotiate with other cloud service providers. Previously, he had pressured Noelle Walsh, who was responsible for data center construction, to shorten the construction timeline.

According to attendees, during a meeting, Altman asked Walsh why she could not expedite the process. Walsh responded that uncontrollable factors, such as uneven concrete pouring by contractors leading to rework, had delayed construction. Altman then questioned whether he should directly raise the issue with Microsoft CEO Satya Nadella.

Microsoft CFO Amy Hood and her team informed colleagues that meeting OpenAI's demands could lead to Microsoft overbuilding servers, posing a risk of not generating financial returns. Azure executives privately believed that constructing several gigawatts of capacity as requested by OpenAI within the specified timeframe was physically impossible.

Ultimately, Nadella and Altman reached a consensus in the summer of 2024 that Microsoft could not be OpenAI's sole cloud service provider. Microsoft then began approving OpenAI's contracts with other suppliers while retaining priority supply rights—allowing it to choose to provide more computing power or permit OpenAI to seek other sources.

Oracle Becomes the Biggest Beneficiary

After obtaining Microsoft's permission, OpenAI quickly took action. In July 2024, the company reached an agreement with Oracle to build a 1-gigawatt data center in Texas. Seven months later, both parties announced an additional agreement for 4.5 gigawatts of capacity, equivalent to the power consumption of two Hoover Dams According to informed sources, OpenAI plans to invest hundreds of billions of dollars in Oracle by the early 2030s. The company has also signed a $22.4 billion cloud contract with CoreWeave, an independent agreement with Google, and agreed to receive up to $100 billion in investments from NVIDIA to build its own data centers.

Some Microsoft executives question whether Oracle can provide an additional 4.5 gigawatts of capacity by 2032 and believe that the agreement may exceed budget since NVIDIA has yet to determine the prices for the next generation of chips. Technology industry analysts and investors have recently expressed similar doubts publicly.

Microsoft's Considerations

Despite losing some business, Microsoft executives believe they made the right decision. Microsoft continues to receive a 20% share of the soaring revenue from OpenAI and has the right to use OpenAI models for free, such as in Microsoft Copilot software. Additionally, Microsoft remains the only cloud partner providing the supercomputers needed for training models for OpenAI.

Hendi Susanto, portfolio manager at Gabelli Funds, stated that this allows Microsoft to better manage the fluctuations in AI demand. Both Microsoft and Oracle are mature large software companies, and free cash flow is a major driver of their performance. I believe Microsoft has a better capital expenditure strategy.

Joe Tigay, portfolio manager at Equity Armor Investments, mentioned that it would be better if OpenAI only used Azure, and I won't lie about that. The two funds he manages are worth over $60 million, with 2.8% in Microsoft shares.

According to those involved in the negotiations between Microsoft and OpenAI, when the agreement was signed in 2023, "every number involved in this deal was larger than we had anticipated at the time. Now we have to figure out how to adjust because the world is changing too fast."

OpenAI's Ambitions for Computing Power Continue to Expand

OpenAI's demand for computing power is still accelerating. Altman recently told colleagues that the company's goal is to use 250 gigawatts of data center capacity by 2033, which is equivalent to one-third of the current peak energy usage in the United States.

According to attendees, during a meeting, Microsoft executive Jon Tinter asked Altman what Microsoft should do if OpenAI decided it needed hundreds of gigawatts of data center capacity—an apparently absurd goal that would require trillions of dollars in investment. The OpenAI team responded seriously: Microsoft could choose to build that capacity or make way for cloud service providers willing to provide it.

According to those who heard Altman speak, he once told another AI researcher that Microsoft's refusal to build new data centers quickly enough is the biggest obstacle to OpenAI's successful development of general artificial intelligence.

According to those involved in the negotiations with Microsoft, "There are people in both companies who hate this. There are people inside Microsoft who don't like it, and there are people inside OpenAI who don't like it. They have many reasons to dislike it... but you still have to solve the problem. Were there tense moments in the conversations? Of course. That's normal." "