What is "cunning and shrewd" Microsoft thinking while watching OpenAI support Oracle?

Over the past year, Microsoft and OpenAI have engaged in a difficult struggle over data center supply. This conflict ultimately ended with Microsoft proactively relaxing its exclusive agreement—OpenAI was allowed to sign large cloud service contracts with competitors such as Oracle, with expected payments to these rivals exceeding what Microsoft itself would receive by 2030. On the surface, it seems that Microsoft is handing over billions of dollars in orders, but the tech giant's calculations are far more astute than outsiders might think.

According to an analysis by tech media The Information, this shift is not a sign of weakness from Microsoft. Insiders revealed that Microsoft Chief Financial Officer Amy Hood and her team repeatedly warned that blindly meeting OpenAI's endless expansion demands could lead to overbuilding, putting Microsoft at risk of not recouping its investments. Meanwhile, Microsoft CEO Satya Nadella remains highly alert to external accusations of Microsoft "controlling" OpenAI—investigations by the Federal Trade Commission and EU regulators are ongoing.

Microsoft's concessions have yielded tangible benefits. Despite relaxing the exclusive terms, Microsoft remains the only cloud service provider offering supercomputers for model training to OpenAI, which is the most critical aspect of AI development. At the same time, Microsoft continues to take a 20% cut from OpenAI's soaring revenues and uses its model technology for its own Copilot products for free. OpenAI plans to pay Microsoft approximately $135 billion in server leasing fees by 2030.

Some Microsoft executives privately question whether Oracle can fulfill its commitments—providing 4.5 gigawatts of data center capacity by 2032, which may be physically challenging to achieve. Similar doubts have also emerged among investors. In contrast, Microsoft's cautious approach to capital expenditures may align better with the financial discipline of a mature tech company.

From Honeymoon to Rift: Exclusive Agreement Becomes a Bottleneck

The tension between Microsoft and OpenAI began in early 2024. At that time, Microsoft was building multiple dedicated data centers for OpenAI, including the "Fairwater" supercomputer project in Wisconsin, costing billions of dollars and consuming nearly 1 gigawatt of power. Microsoft planned to have the Wisconsin and Atlanta data centers operational by 2026 and to launch a second Wisconsin facility by 2028.

However, OpenAI CEO Sam Altman believed this was far from sufficient. He demanded that Microsoft provide several gigawatts of additional computing power by the end of the century, which would require Microsoft to invest hundreds of billions of dollars in purchasing NVIDIA chips and power infrastructure. According to a participant, Altman once told an AI researcher that Microsoft's refusal to accelerate data center construction was the "biggest obstacle" to OpenAI achieving general artificial intelligence.

Engineering delays exacerbated the conflict. The Wisconsin project was delayed by several months due to uneven concrete pouring issues with contractors. During a meeting in mid-2024, Altman questioned Microsoft executive Noelle Walsh, who was responsible for data center construction, about why the timeline could not be shortened, even suggesting escalating the issue to Nadella's level. According to attendees, Walsh explained that the delays were due to construction factors beyond Microsoft's controlMicrosoft has its own considerations. Hood's finance team is concerned that unlimited expansion according to OpenAI's demands may lead to resource waste. Microsoft Azure executives privately believe that no cloud service provider can deliver such massive computing power within the timeframe requested by OpenAI—this judgment later became an important basis for Microsoft's agreement to lift the exclusive agreement.

Breakthrough and Compromise: Allocation of the $450 Billion Budget

The summer of 2024 became a turning point. Altman requested Microsoft to allow him to negotiate with other cloud service providers, while Nadella was quite sensitive to public accusations of "Microsoft stifling OpenAI." Under Nadella's direction, Hood's team reached a new agreement with OpenAI's Chief Financial Officer Sarah Friar: Microsoft retains priority supply rights and can choose to meet OpenAI's demands or let it seek other suppliers.

Subsequently, OpenAI acted quickly, signing its first 1-gigawatt Texas data center contract with Oracle in July 2024, and in December received Microsoft's approval to conduct a second round of negotiations, ultimately announcing a massive 4.5-gigawatt agreement seven months later. According to an OpenAI employee, the company has allocated a budget of $450 billion for server expenditures before 2030, with payments to competitors like Oracle expected to exceed those to Microsoft.

The scale of these agreements is staggering. In addition to Oracle, OpenAI also signed a $22.4 billion cloud services contract with CoreWeave, rented servers from Google Cloud, and plans to utilize NVIDIA's investment of up to $100 billion to build its own data centers. Oracle's stock surged 36% after announcing that its cloud business would generate $400 billion in revenue over the next four years, with OpenAI orders being a major driving force.

However, Microsoft has not completely exited; in addition to the expected revenue of $135 billion, Microsoft still exclusively provides the supercomputers needed for model training—this is the core aspect of AI development, far more critical than inference deployment. A person involved in Microsoft's negotiations stated, "Every number is larger than we anticipated when we signed in 2023; now we need to figure out how to adapt to a rapidly changing world."

Shrewd Retreat: Microsoft's Risk Management Logic

Microsoft shareholders have differing opinions on this arrangement. Joe Tigay, portfolio manager at Equity Armor Investments, admitted, "It would certainly be better if OpenAI only used Azure; I won't lie." However, some investors recognize Microsoft's cautious strategy.

Hendi Susanto, portfolio manager at Gabelli Funds, pointed out, "This allows Microsoft to better manage the fluctuations in AI demand; we can expect imbalances. Both Microsoft and Oracle are mature large software companies, and free cash flow is the main driver of performance. From this perspective, Microsoft's capital expenditure strategy is superior."

Internally, Microsoft is skeptical about whether Oracle can fulfill its commitments. Two insiders revealed that some executives believe it may be "physically impossible" for Oracle to deliver 4.5 gigawatts of capacity before 2032, and the project may exceed budget due to the next generation of NVIDIA chips not yet being priced. In recent weeks, several technology industry analysts and investors have publicly expressed similar doubts.**

At a meeting, Microsoft executive Jon Tinter once asked Altman: What should Microsoft do if OpenAI needs hundreds of gigawatts of capacity—an objective that would cost trillions of dollars? According to attendees, the OpenAI team responded seriously: Microsoft could choose to build it or let suppliers willing to do so take over. This almost arrogant ambition made Microsoft more convinced that rather than taking on astronomical financial risks, it was better to let competitors take the gamble.

A person involved in the Microsoft negotiations admitted that there were people within both companies who were dissatisfied with the partnership, but "those who really understand the situation" know this is the only viable solution. "Were there tense moments in the discussions? Of course," the person said, "but you still have to find a way to solve the problems." Altman recently told colleagues that OpenAI plans to use 250 gigawatts of data center capacity by 2033—equivalent to one-third of the current peak energy usage in the United States. Faced with such appetite, Microsoft's concession may not be weakness, but a form of pragmatic self-preservation