OpenAI is reportedly preparing for a large change in the way it sources the computing power behind its AI systems. According to The Information, the company expects Stargate to become its dominant source of data center capacity by 2030.
The shift would not be immediate. OpenAI still plans to increase spending on Microsoft-owned data centers in the next few years, even as its longer-term forecast points toward a much larger role for Stargate, a project expected to be heavily financed by SoftBank.
What OpenAI reportedly expects by 2030
The central forecast is straightforward: by 2030, OpenAI expects to get three-quarters of its data center capacity from Stargate. That would represent a major change in the company’s compute mix because Microsoft fulfills most of OpenAI’s power needs today.
Microsoft is also OpenAI’s biggest shareholder, which makes the reported plan especially significant. It does not describe an abrupt break with Microsoft. Instead, it suggests a staged transition in which OpenAI keeps relying on Microsoft-owned data centers while preparing for Stargate to take on a far larger share of capacity later.
SoftBank is one of OpenAI’s newest financial backers, and Stargate is expected to be heavily financed by SoftBank. If the forecast plays out as described, SoftBank’s role would move beyond financial backing and into the infrastructure layer that supports OpenAI’s AI model operations.
Why data center capacity matters
For an AI company, data center capacity is not a background detail. It is part of the core operating base. Training AI models and running AI models both depend on access to large amounts of computing power, and the source of that capacity can shape costs, planning, and strategic flexibility.
The reported forecast shows OpenAI looking several years ahead at how much infrastructure it may need and where that infrastructure may come from. The company’s products and model operations require power today, but the plan described by The Information is about what the company expects its capacity needs to look like over the next five years.
That is why the expected move toward Stargate matters. If one project is forecast to provide three-quarters of OpenAI’s data center capacity by 2030, it would become central to how the company supports its AI systems. The scale of that share is what makes the reported plan more than a routine vendor adjustment.
The Microsoft relationship is still expanding first
The report does not say OpenAI is reducing Microsoft spending right away. In fact, it says the opposite for the near term: OpenAI still plans to increase its spending on Microsoft-owned data centers in the next few years.
That detail is important because it frames the Stargate forecast as a longer-term rebalancing. Microsoft remains the company that fulfills most of OpenAI’s power needs today. The reported change is about where the majority of capacity is expected to come from by 2030, not about an overnight replacement.
This creates a two-track picture. In the short run, OpenAI’s spending on Microsoft-owned data centers is set to rise. In the longer run, OpenAI reportedly expects Stargate to become the much larger source of data center capacity.
Costs are expected to rise sharply
The forecast also points to much higher spending. The Information reports that OpenAI projects to burn $20 billion in cash during 2027. That is far more than the $5 billion it reportedly burned through in 2024.
Those figures underline the financial pressure behind the infrastructure plan. More compute capacity can support more AI activity, but it also brings major cost demands. The reported numbers suggest that OpenAI is planning for a period in which its operating needs grow dramatically.
The report also says that by 2030, OpenAI forecasts that costs around running AI models, also known as inference, will outpace what it spends on training AI models. That distinction matters because training and inference are different parts of the AI lifecycle.
Training is the work of creating or improving models. Inference is the work of running those models when they are used. If inference costs are expected to exceed training costs by 2030, then ongoing usage becomes a larger cost center than the model-building phase itself.
What the reported shift signals
The reported plan signals that OpenAI is thinking about infrastructure as a long-term strategic issue. The company’s future costs are tied not only to building AI models, but also to keeping them running at scale.
It also shows how financial backing and computing infrastructure can become closely connected. SoftBank is described as one of OpenAI’s newest financial backers, while Stargate is expected to be heavily financed by SoftBank. If Stargate becomes the source of three-quarters of OpenAI’s data center capacity by 2030, that relationship could become deeply important to OpenAI’s operating model.
At the same time, Microsoft remains central in the current picture. OpenAI’s biggest shareholder fulfills most of its power needs today, and OpenAI still plans to increase spending on Microsoft-owned data centers in the next few years. The forecast is not a simple exit from Microsoft, but a reported move toward a different balance of compute supply over time.
The larger takeaway is that OpenAI’s AI ambitions appear to require a much larger infrastructure base, much higher spending, and a changing mix of partners. If the company’s projections hold, Stargate will become a defining part of that compute strategy by 2030.