OpenAI raised its five-year revenue forecast by roughly 27% and told investors it plans to spend about $600 billion on compute through 2030, according to The Information, which obtained slides the company shared ahead of a funding round that could exceed $100 billion. The company pulled in $13.1 billion in revenue in 2025, beating its own $10 billion target. But the real story is buried further down in the deck: adjusted gross margin fell to 33%, down from 40% the year before.
That margin number deserves a pause. Just weeks ago, OpenAI was touting 70% compute margins on paid products. The 33% figure is an adjusted gross margin, defined as revenue minus inference costs, and it came in well below the 46% the company had targeted for 2025. The culprit, per the slides: OpenAI had to scramble for more expensive compute capacity when demand outstripped expectations. Beating your revenue target while missing your margin target is a specific kind of problem, the kind that gets more expensive to fix the faster you grow.
The numbers that matter
OpenAI now expects 2026 revenue to hit $30 billion, roughly doubling from 2025, with 2027 reaching $62 billion. By 2030, the company projects more than $280 billion in total revenue, split roughly evenly between consumer and enterprise. CFO Sarah Friar recently said annualized revenue topped $20 billion by the end of 2025.
Consumer sales are expected to double to $17 billion this year and climb to $150 billion by 2030, representing just over half of total revenue at that point. Revenue from new products, including hardware sales, is projected at $1.3 billion next year. What exactly falls under that category beyond hardware remains unclear.
These are the numbers OpenAI wants investors to see right before writing very large checks. Nvidia is negotiating to invest up to $30 billion as part of a round that could value OpenAI at $730 billion pre-money, with SoftBank and Amazon also participating. When a company updates its projections weeks before a mega-round, the timing is the message.
Where the money goes
The spending side is staggering, even by OpenAI's standards. The $600 billion compute target through 2030 is actually a reduction from CEO Sam Altman's previous boast of $1.4 trillion in infrastructure commitments. Framing a $600 billion spend plan as fiscal discipline tells you something about the world OpenAI operates in.
In 2025, the company spent more than $8 billion running its AI models for users, with roughly $4.5 billion going to inference for paying customers. That figure is expected to roughly double to $14 billion this year and hit $26 billion in 2027. Training costs are even steeper: $8.3 billion in 2025 (about a billion under the summer forecast), ballooning to $32 billion this year and $65 billion next year. Through 2030, training alone is projected to cost nearly $440 billion.
Add it up and OpenAI expects to spend $25 billion this year and $57 billion next, collectively about $30 billion more than previously forecast. The company says it will burn more than double the cash it had previously projected through 2030, though it still insists it will reach positive cash flow by then. OpenAI had roughly $40 billion in cash at the end of 2025, which sounds like a lot until you look at the spending trajectory.
So does anyone care?
ChatGPT hit a new peak of more than 910 million weekly active users, according to the slides. OpenAI's leadership told investors that weekly active user growth slowed in early fall, a detail that probably caused some discomfort in the room. Subsequent releases of GPT-5.1 and GPT-5.2, which the company says made the model more "human" and improved overall performance, reignited growth. The company projects 2.75 billion WAU by 2030. Its coding tool, Codex, has passed 1.5 million weekly active users.
The 910 million figure is impressive, but context matters. Most of those users are on the free tier. Converting free users into paying customers, or into advertising revenue that justifies serving them at a loss, is the entire business model question. OpenAI recently started testing ads for some users. Whether nearly a billion weekly users translates into $280 billion in annual revenue by 2030 depends on assumptions about willingness to pay that no one can verify yet.
The 2.75 billion WAU target for 2030, meanwhile, would represent roughly a third of the world's population. For reference, Meta needed 20 years and multiple product lines to reach 3.3 billion monthly active users. OpenAI is projecting something in that neighborhood within about five years for a single product family.
Reading between the slides
There is a tension running through these numbers that the slides don't resolve. Revenue is growing fast, but margins are shrinking. The company beat its revenue target in 2025 while spending less than expected on inference ($8 billion versus a $9 billion target, per Reuters), yet gross margins still fell. That math only works if revenue mix shifted toward lower-margin products or if the emergency compute purchases were expensive enough to overwhelm efficiency gains elsewhere.
OpenAI's December "code red," prompted by Google's Gemini gaining ground, is also worth considering. The company redirected resources to improve ChatGPT and delayed revenue initiatives like advertising and shopping agents. That kind of defensive scramble does not usually coincide with the smooth upward projections presented to investors.
The Musk trial, set for March 30, looms as a wildcard. Unsealed documents have already produced uncomfortable quotes from co-founder Greg Brockman's diary. OpenAI values its exposure at $38 million; Musk is seeking up to $134 billion.
OpenAI's funding round is expected to close in the coming weeks, with a potential IPO filing in the second half of 2026.
