Your favorite AI company just admitted it burned through $34 billion last year and still only makes $13 billion in revenue. The math does not get better from here.
OpenAI's audited financials leaked on Monday via Ars Technica and Fortune, and the numbers tell a story that neither the company nor its investors wanted public right now. The company is preparing for an IPO, and these documents arrived at the worst possible moment.
What the Numbers Actually Show
OpenAI's revenue tripled from $3.7 billion in 2024 to $13.07 billion in 2025. That sounds impressive until you look at the other side of the ledger. Total costs hit $34 billion. That means for every dollar OpenAI earned, it spent $2.60.
Here is the full picture:
| Metric | 2024 | 2025 |
|---|---|---|
| Revenue | $3.7B | $13.07B |
| R&D Expenses | $7.81B | $19.18B |
| Cost of Revenue | $2.65B | $7.5B |
| Sales & Marketing | $1.11B | $5.73B |
| Operating Loss | $8.78B | $20.92B |
| Total Expenses | $12.48B | $34B |
The R&D line is the one that should worry you. OpenAI paid Microsoft $10.59 billion alone for R&D-related expenses in 2025. That is not "research and development" in the traditional sense. It is computeleasing at industrial scale, repackaged as an R&D line item.
The net loss headline is $38.5 billion, but that includes a one-time $30 billion accounting charge tied to the company's conversion from a nonprofit to a for-profit structure. Strip that out and the real operating loss is roughly $8.5 billion. Still massive, but a different conversation.
The Efficiency Story OpenAI Wants You to Focus On
The company will point you to one number: it cost $2.37 to generate $1 of revenue in 2024. By 2025, that dropped to $1.60. Revenue is growing faster than costs. The gap is narrowing.
This is the narrative that matters for the IPO. OpenAI is not trying to convince you it is profitable today. It is trying to convince you the trajectory bends toward profitability by 2030, which is the timeline the company has communicated to investors.
Monthly revenue hit nearly $2 billion by the end of 2025. If that run rate holds, 2026 revenue could approach $24-25 billion. But the question is whether costs grow in lockstep.
Who Is Actually Paying
Here is the uncomfortable part: OpenAI reports 900 million weekly active users. Only 50 million are paying subscribers. That is a 5.5% conversion rate. The other 895 million people are using ChatGPT for free, which means OpenAI is subsidizing their usage with investor capital.
Enterprise customers are also pushing back. A Forbes survey from January found 56% of CEOs see zero ROI from their AI spending. GitHub Copilot users are reacting badly to new usage-based pricing. The customers who actually pay are starting to demand proof that the product is worth the bill.
The Competitive Pressure
Anthropic is not just competing on model quality. It is creating downward pricing pressure across the industry. When Claude performs comparably to GPT-5 at a lower price point, OpenAI cannot maintain premium pricing without losing market share.
Meanwhile, open-source models like Qwen, DeepSeek, and the new GLM 5.2 are eating the bottom of the market. Companies that were paying OpenAI for API access are discovering they can run capable models locally for a fraction of the cost.
OpenAI has already started cutting side projects to manage costs. The Sora video model was shut down in March 2026. The company is narrowing its focus to core products and coding tools, which suggests management knows the current burn rate is not sustainable across a broad product portfolio.
The Cash Question
OpenAI raised $122 billion in March 2026 and holds $25 billion in cash. At the current burn rate, that gives the company roughly 18 months of runway before it needs to raise again or show a path to profitability.
The $852 billion valuation means investors are pricing in a future that does not exist yet. If the IPO happens at that valuation, OpenAI needs to demonstrate that the efficiency trend continues and that revenue can eventually outpace the $19 billion annual R&D bill.
What Surprised Me
The number that caught my attention is the $5.73 billion in sales and marketing. That is a 5x increase from 2024. OpenAI is spending more on selling the product than on running it (cost of revenue was $7.5B, but a chunk of that is also Microsoft payments). This looks like a company that knows its organic growth is stalling and is buying its way to revenue targets.
The 900 million user figure is a vanity metric if only 5.5% pay. Spotify converts about 40% of free users to premium. Netflix does not even have a free tier. OpenAI's conversion rate suggests the product is useful enough to try but not compelling enough to buy.
The HN discussion was split. Some commenters argued R&D-heavy spending is normal for a capital-intensive industry and compared the trajectory to Uber's path to profitability. Others pointed out that extrapolating the current loss curve implies $45 billion in losses this year and $90 billion next year. Neither side is wrong. The question is whether the efficiency improvements are structural or just the company deferring costs it cannot eliminate.
My take: OpenAI is not going bankrupt. It has enough cash and investor goodwill to survive through 2027. But the era of "spend whatever it takes" is over. The side project cuts, the pricing pushback, and the leak itself all point to a company that is being forced to grow up faster than it wanted to. The IPO will force transparency, and transparency will force discipline.