In June 2026, a leaked OpenAI financial document sent shockwaves through the tech community. The document revealed that OpenAI's revenue in 2025 reached $13.07 billion, a staggering 253% increase from $3.7 billion in 2024. However, accompanying the soaring revenue were operating losses of $20.92 billion, with a net loss of approximately $8 billion.
Behind the prosperous facade of ChatGPT exceeding 900 million weekly active users and a company valuation of $852 billion, OpenAI's ledger exposes a harsh reality: in 2025, for every $1 the company earned, it spent $1.6. Is this 'burning money for scale' model a unique pain point for OpenAI on its path to Artificial General Intelligence (AGI), or is it a common ailment across the entire large model industry? By dissecting its cost structure and comparing it horizontally with financial data from leading companies like Anthropic and xAI, we might get a clearer picture of the true cost behind the current AI industry boom.
The Cost Black Hole Behind $13 Billion in Revenue: Where Did the Money Go?
To understand the logic behind OpenAI's losses, we must first dissect the composition of its $34 billion in total costs and expenses. In this leaked financial document, the largest expense item was R&D costs, amounting to $19.18 billion, which included a payment of $10.59 billion to Microsoft. This is followed by $7.5 billion in cost of revenue (primarily for inference computing) and $5.73 billion in sales and marketing expenses.
From a growth perspective, OpenAI's money-burning efficiency actually improved. In 2024, the company spent $2.37 for every $1 of revenue generated, while by 2025, this figure dropped to $1.6. Revenue growth (253%) outpaced total cost growth (172%). However, this does not mean the cost pressure has eased. On the contrary, the price of admission dictated by the scaling law is still rising sharply.
The $19.18 billion in R&D expenditure accounted for a whopping 147% of its annual revenue. In the large model field, R&D signifies not just algorithm engineers' salaries but, more importantly, massive training compute consumption. To maintain a lead in model capabilities, OpenAI must continuously invest heavily in training the next-generation models. This investment is rigid; once it slows down, it risks losing its position in the race against competitors.
The $7.5 billion inference computing cost is equally significant. This cost is directly tied to user usage volume. With ChatGPT exceeding 900 million weekly active users, a massive number of inference requests flood OpenAI's servers daily. Every conversation, every generation consumes real computing resources. Although hardware performance improves, user demand for more complex, longer-context interactions grows even faster, causing the absolute value of inference costs to continue climbing.
Furthermore, the $5.73 billion in sales and marketing expenses reflects the high cost for AI companies in acquiring C-end customers and expanding in the enterprise sector. As product homogenization begins to emerge, maintaining brand visibility and capturing enterprise client share requires substantial financial investment.
It's crucial to clarify the net loss metric. The leaked document shows that the 2025 net loss included a one-time, non-cash accounting expense of approximately $30 billion. This stemmed from fair value changes of convertible equity and warrant liabilities when OpenAI transitioned from a non-profit structure to a Public Benefit Corporation (PBC). Excluding this one-time factor, the actual operational loss was about $20.92 billion, with a net loss of roughly $8 billion. This distinction is essential as it removes the book fluctuations caused by the financial structure change, revealing the real consumption of the company's daily operations.
A $17.2 Billion Structural Burden: Microsoft's 'Invisible Take'
Within OpenAI's cost structure, there is an unavoidable behemoth: Microsoft. According to the leaked document, OpenAI paid Microsoft a total of $17.2 billion in 2025. This included $10.59 billion in R&D expenses, $6.047 billion in cost of revenue, $527 million in sales expenses, and $42 million in administrative expenses.
This $17.2 billion payment accounted for 50.5% of OpenAI's total annual costs, even exceeding its $13.07 billion annual revenue. Microsoft is not just OpenAI's cloud service provider; it is also an 'invisible shareholder' deeply tied to OpenAI's cash flow through compute revenue sharing. In the early stages of cooperation, Microsoft's compute support was key to OpenAI's rapid rise. However, as OpenAI's business scaled, this sharing model evolved into a heavy structural burden.
According to previously disclosed cooperation agreements, OpenAI must pay Microsoft a 20% revenue share until 2030. This means that as long as OpenAI uses Microsoft's Azure cloud services for training and inference, this expenditure will persist. Before achieving positive cash flow, OpenAI must first cover Microsoft's compute bill. This structure also explains why OpenAI needed to complete a massive $122 billion financing round in March 2026. When self-generated cash flow is insufficient, external funding is the only way to maintain operations.
Money-Burning Efficiency Ranking: OpenAI vs. Anthropic vs. xAI
Is high R&D and high loss unique to OpenAI? Turning our gaze to two other leading AI companies, the answer is no.
According to SpaceX's submitted IPO S-1 filing, Elon Musk's xAI had revenue of $3.2 billion in 2025, but operating losses reached $6.4 billion, with capital expenditures even hitting $12.7 billion. Calculating money-burning efficiency, xAI spends $3 for every $1 earned, with a loss/revenue ratio of 200%, far higher than OpenAI's 160%. To bet on trillion-parameter models, xAI built the Colossus data center in just 122 days, with capital expenditures exceeding the combined capex of SpaceX's Starlink and rocket businesses. This indicates that on the pursuit of the scaling law track, xAI is placing a more extreme, asset-heavy bet than OpenAI.
The situation with another major competitor, Anthropic, presents a different path. According to official announcements, Anthropic's annualized revenue (ARR) reached $9 billion by the end of 2025 and skyrocketed to $47 billion by May 2026. Its core growth engine, Claude Code, had an ARR exceeding $2.5 billion by February 2026.
However, rapid growth also conceals cost pressure. As reported by The Information, Anthropic's gross margin in 2025 was only 40%, 10 percentage points lower than expected, due to inference costs being 23% higher than anticipated. Regarding losses, media reports suggest its EBITDA loss is also in the tens of billions. Lacking precise audit documents, we cannot know Anthropic's actual total net loss, but the 40% gross margin and higher-than-expected inference costs expose the same industry-wide pressure.
Comparing data from the three companies side-by-side reveals that in 2025, the combined operating losses of OpenAI, xAI, and Anthropic exceeded $30 billion. Burning money for scale is not an isolated case; it is the norm in the current large model competition. The difference lies in the choice of commercial path. Anthropic does not build its own data centers, relying on a multi-cloud strategy with AWS, Google, and Azure, taking a light-asset route, and achieving high-premium monetization through Claude Code in the enterprise sector. xAI firmly controls its compute infrastructure, betting on compute monopoly. OpenAI sits somewhere in between, relying on Microsoft's compute while possessing a massive C-end user base.
900 Million Weekly Actives & 5.6% Conversion Rate: Stress Testing the Monetization Ceiling
A massive user base is OpenAI's core moat and a key support for its $852 billion valuation. However, the financial data reveals the other side of this moat.
Among ChatGPT's 900 million weekly active users, paying users number approximately 50 million, a conversion rate of about 5.6%. Roughly estimating based on $13.07 billion in revenue, the annual revenue per paying user (ARPU) is about $261. This means over 800 million free users are consuming compute resources without generating direct revenue.
Against the backdrop of persistently high inference costs, the compute consumption by free users becomes a massive burden. How to increase the conversion rate and ARPU is a direct challenge facing OpenAI. Compared to Anthropic's strategy, this pressure is even more apparent. Facing cost pressure, Anthropic chose to double the price of its top-tier model API and introduced tiered pricing strategies like Claude Fable, turning top-tier AI capabilities into 'luxury goods' to filter for high-value enterprise clients.
OpenAI, however, still maintains its basic $20-per-month subscription model. This model aids rapid user base expansion during the growth phase, but during a stage requiring cost structure optimization, it inevitably faces pressure to raise prices or implement further tiered pricing.
Who Foots the Bill for the Scaling Law?
OpenAI's leaked ledger tears open a corner of the AI industry's glamorous exterior. Earning tens of billions annually while losing billions is not only OpenAI's current state but also a dilemma shared by leading companies like xAI and Anthropic. High R&D investment and high inference costs constitute the two major mountains in large model competition.
Massive funding rounds provide a cushion for this money-burning model. The $122 billion financing completed by OpenAI in March 2026 and Anthropic's valuation reaching $965 billion in May the same year indicate that capital markets are still willing to pay for the scaling law—for now. But capital's patience is limited.
Whether AI companies can escape the loss quagmire depends on achieving a drastic reduction in marginal costs. Early-stage SpaceX slashed launch costs by over 90% through rocket reusability, transforming the economics of the aerospace industry. Whether the AI industry can replicate this path depends on whether inference compute costs can be drastically reduced through specialized chips, model compression, or architectural innovation. Until then, high R&D and high losses will remain the dominant theme of the AI industry. What determines whether AI tools can continue to evolve is not the brilliance of the algorithms, but the cost structure hidden in the ledgers.
