OpenAI is preparing for an AI future in which access to computing power may matter as much as model design, product strategy, or distribution. Its Stargate data center push shows how far the company is willing to go to secure that capacity, even as Microsoft remains a major partner and investor.
The result is not a simple split. It is a change in balance. OpenAI appears to be moving faster and taking on more infrastructure risk, while Microsoft is expanding with more caution.
Stargate moves to the center of OpenAI's compute strategy
According to investor documents reviewed by The Information, OpenAI plans to get about 75% of its computing power from the "Stargate" data center project by 2030. Stargate is described as a joint venture with SoftBank and Oracle.
If that plan holds, Stargate would become the dominant source of OpenAI's infrastructure. That would effectively move Microsoft out of the role of primary infrastructure partner, even though Microsoft would remain financially tied to OpenAI's success.
The scale is central to the story. OpenAI aims to build data centers with 8 gigawatts of total capacity. The source describes that level as comparable to the entire data center operations of major cloud providers.
For an AI company, that kind of capacity is not just back-end plumbing. It shapes how many models can be trained, how quickly products can serve users, and how much room the company has to absorb demand spikes. OpenAI's plan suggests it sees future AI competition as a race for physical infrastructure as well as software.
Microsoft spending is still set to rise
The Stargate plan does not mean OpenAI is cutting Microsoft out of its business. Current contracts show OpenAI's Microsoft expenditure rising from $13 billion this year to $28 billion in 2028, according to The Information.
That spending could increase further if Microsoft supplies more capacity. Microsoft also remains a major investor, which means OpenAI's growth can still benefit Microsoft financially.
This is why the relationship is more complicated than a clean replacement. OpenAI is trying to add another large source of computing power, while Microsoft is still deeply embedded in the economics of the company.
The tension is about timing and risk. OpenAI appears to be acting as if it must secure as much capacity as possible before demand and competition outpace available supply. Microsoft, by contrast, is described as taking a more structured route to expansion.
The cost of chasing capacity
OpenAI's growth projections are large, but so are the expected losses tied to its compute ambitions. The company projects revenue will more than triple from $3.7 billion this year to $12.5 billion, with a target of $28 billion by 2026.
SoftBank companies are expected to account for one-third of this year's growth through usage of OpenAI products. That link matters because SoftBank is also tied to the broader Stargate financing story.
The spending side is even more aggressive. OpenAI's cash burn is expected to rise from about $2 billion last year to nearly $7 billion this year, then peak at around $20 billion in 2027. Profitability is not expected until the end of the decade.
Much of that pressure comes from computing investments. OpenAI plans to spend over $320 billion on AI model training and operations between 2025 and 2030. Half of that amount is allocated to training, while operating costs are projected to exceed training costs after 2030.
The financing plan is also large. OpenAI is preparing a $40 billion financing round, with SoftBank potentially contributing $30 billion. The round would value OpenAI at $260 billion, with the first $10 billion targeted by March's end. The Stargate project's total investment is estimated at $500 billion.
Microsoft sees a softer market ahead
Microsoft CEO Satya Nadella is presented as more cautious on future compute investment. He expects a surplus of computing power by 2027/2028, which could push prices lower. His position is summed up by the phrase "happy to lease" if the market softens.
Microsoft is not standing still. It is following a structured expansion plan with an $80 billion data center expansion budget for fiscal year 2025.
The difference is not whether AI needs more infrastructure. Both companies are preparing for that. The difference is how much capacity to secure now, how much to build directly, and how much risk to take before demand is fully proven.
OpenAI's own usage figures help explain why it may prefer speed. The company recently reported over 400 million weekly active users, an 8% increase in recent weeks alone. API usage has reportedly grown fivefold since introducing o3-mini, the company said.
More product launches could add to that pressure. GPT-4.5 will reportedly be released in the coming days, with GPT-5 planned for around May.
Competition adds urgency
OpenAI's concern is not only internal demand. The company is also watching competitors move quickly.
OpenAI started working with Oracle for additional computing power in summer 2024. By October, CEO Sam Altman reportedly worried that Microsoft could not deliver servers fast enough to match Elon Musk's xAI, which just released its Grok 3 model.
The source points to xAI's infrastructure buildout as part of that pressure. Musk's Memphis data center already houses 200,000 Nvidia chips and continues to grow. A second facility under construction in Atlanta has 12,000 GPUs installed so far.
Taken together, these details show why Stargate matters. OpenAI is trying to reduce dependence on one infrastructure path while keeping Microsoft in the picture. It is also betting that future AI demand will justify extraordinary spending on data centers, chips, training, and operations.
That bet may define the next stage of the AI market. If OpenAI is right, the companies with the largest and fastest computing pipelines could set the pace. If Microsoft is right about a coming surplus, aggressive early builders may face a different economics problem: owning or funding capacity in a market where leasing becomes cheaper.