OpenAI posted a net loss of about $38.53 billion in 2025, a brutal jump from about $5.09 billion in 2024, according to audited financial documents reviewed by the Financial Times. Revenue did rise to $13.07 billion, but expenses still outpaced it by a wide margin. The numbers show a company spending like a hyperscaler while still trying to prove its public-market story, which is exactly the kind of tension that makes an OpenAI IPO filing more interesting than any glossy product demo.
That gap leaves OpenAI with a business that is growing fast and burning even faster, which is not an uncommon pattern in frontier AI – but it is a lot less charming once the invoices get this large.
R&D and Microsoft payments drove the bill
The biggest cost line was research and development at $19.18 billion, followed by sales and marketing at $5.73 billion and administrative expenses at $1.57 billion. Total operating loss reached $20.92 billion, while overall costs came to roughly $34 billion. That kind of spend is what happens when a company is simultaneously building models, scaling infrastructure, and trying to own the market before rivals can get a firmer grip.
Microsoft was a major part of the picture. OpenAI paid the company about $10.59 billion for R&D services, while related obligations reached $17.2 billion; Microsoft also contributed about $303 million in payments to OpenAI. SoftBank added about $867 million in investment funding, underscoring how much outside capital is still propping up the machine.
A balance sheet padded by cash, not comfort
OpenAI ended the year with assets above $50 billion, with a significant share sitting in cash. That gives the company room to keep spending, but it does not erase the basic problem: costs are rising faster than revenue, and the math gets harsher as the business structure becomes more public-facing and more expensive to defend.
The financial documents also point to a major structural shift, including a move toward a more commercial model and the revaluation of convertible obligations, which created large accounting adjustments. Translation: part of the damage is operational, and part of it is the messy accounting that comes with preparing a company for a potential IPO.
What the IPO crowd will focus on
The next question is whether OpenAI can keep scaling revenue quickly enough to convince public investors that these losses are a phase, not a feature. If the company heads toward an IPO, the market will likely care less about model hype and more about whether the spend curve can bend before it starts bending the valuation the other way.