OpenAI spent $34 billion in 2025 while losing $39 billion — numbers that would stagger any private company. Leaked financial documents obtained by multiple news outlets reveal the staggering cost of running the world's most ambitious AI lab, and they raise urgent questions about whether the company's planned IPO can succeed under such a massive burn rate.

The leaked OpenAI financials, first reported by Ars Technica, show a company spending over $93 million per day on compute, talent, and infrastructure. With revenue of roughly $13 billion against $34 billion in operating expenses, the gap between income and spending has never been wider — and it is widening every quarter.

Inside the OpenAI Leaked Financials: Where $34 Billion Went

The leaked documents break down 2025 spending into four major categories, each revealing a different facet of the company's financial strategy — and its fragility.

Compute costs: the $18 billion chokepoint

More than half of OpenAI's spending — roughly $18 billion — went to compute infrastructure in 2025. This covers Azure cloud rentals from Microsoft, custom GPU clusters, inference servers for ChatGPT and the API, and the massive training runs for GPT-5-class models. With Nvidia's latest B300 GPUs costing over $30,000 per unit and AI training clusters consuming 50 megawatts or more, compute is by far the single largest expense on OpenAI's balance sheet.

To put this in perspective, OpenAI spends more on compute annually than the entire GDP of several small nations. Each GPT-5-scale training run is estimated to cost between $500 million and $1 billion when factoring in GPU depreciation, power, cooling, and data center staff.

Talent and compensation: the $8 billion war for talent

OpenAI employed roughly 4,500 people by end of 2025, with an average compensation cost exceeding $1.7 million per employee when factoring in stock-based compensation, bonuses, and retention packages. The leaked documents show that OpenAI spent $8 billion on talent costs in 2025 alone — a figure that includes the high-profile defection of CTO Mira Murati and the subsequent retention packages needed to keep key researchers from following Noam Shazeer and other senior departures.

By comparison, Google's DeepMind spends roughly $3.5 billion annually on talent with over 2,000 employees — about half OpenAI's per-employee cost.

OpenAI spending breakdown by category: compute dominates at $18B, followed by talent at $8B.

Infrastructure and data centers: the $5 billion build-out

Beyond compute rentals, OpenAI invested $5 billion in building its own data center capacity and fiber infrastructure. This includes the massive Stargate project partnerships and prepayments for future GPU capacity. These are capital expenditures that will depreciate over time, but they consume cash today and contributed to the $39 billion net loss figure that includes non-recurring charges.

Operating expenses and everything else: $3 billion

The remaining $3 billion covers office space, legal costs (including the ongoing Musk lawsuit and regulatory battles), API partner payouts, marketing, and administrative overhead. OpenAI's legal spending alone hit an estimated $400 million in 2025 as the company fought multiple lawsuits over training data copyright, safety practices, and its non-profit-to-for-profit restructuring.

OpenAI Revenue vs Spending: The Widening Gap

The financial data paints a stark picture of a company racing to scale revenue before its cash runway runs out. While $13 billion in annual revenue sounds impressive — and places OpenAI among the fastest-growing software companies in history — it covers only 38% of the company's annual spending.

Revenue breakdown: API and subscriptions

OpenAI's revenue comes from two primary sources. ChatGPT subscriptions (Plus, Pro, Team, and Enterprise tiers) generated approximately $7.5 billion in 2025, making it the dominant revenue stream. The API business, serving developers and enterprises, brought in roughly $5 billion. A small remainder comes from licensing and partnership deals with companies like Microsoft and Apple.

The challenge: ChatGPT subscription growth is slowing. OpenAI added roughly 200 million weekly active users in 2025, bringing the total to over 600 million, but the conversion rate from free to paid users has plateaued around 4-5%. Meanwhile, API revenue faces increasing price pressure from open-source alternatives like Meta's Llama 4 and Mistral, which offer comparable capabilities at 60-80% lower cost.

Comparing burn rates across AI leaders

No other AI company comes close to OpenAI's burn rate. Anthropic, the second-largest standalone AI lab, operates at roughly one-third the spending level. DeepMind and Microsoft AI have the luxury of parent-company subsidies — OpenAI has no such safety net.

What the OpenAI Leaked Financials Mean for the IPO

The leaked reports land at a critical moment. The company has been preparing for an initial public offering that could value it at over $300 billion, based on the SoftBank-led funding round earlier this year. But the leaked numbers raise fundamental questions about whether public market investors will accept a company losing $21 billion annually on operations.

Timeline uncertainty

Before the leak, OpenAI's leadership had signaled a potential IPO in late 2026. The leaked documents were reportedly prepared as part of confidential IPO roadshow materials, suggesting the company was further along in the process than publicly known. Industry analysts now expect the company to delay its public offering by at least 6 to 12 months to give itself time to demonstrate a credible path to profitability.

Path to profitability: unrealistic at current burn

To reach operating profitability, OpenAI would need to either triple its revenue (to roughly $40 billion annually) or cut spending by more than half. Tripling revenue within 24 months is unprecedented even by OpenAI's growth standards. Cutting spending means reducing either compute capacity (which would hurt model quality and latency) or talent (which would slow research output). The leaked financials show no realistic path to break-even before 2028 at the earliest.

Investor sentiment shift

The leak comes at a time when public market investors are growing skeptical of high-burn AI companies. The failed IPO of CoreWeave in early 2026 and the struggling public debut of several AI infrastructure SPACs have already cooled the market for AI-adjacent listings. OpenAI's leaked financials may push institutional investors to demand stronger profitability commitments than the company can deliver.

The path to IPO: OpenAI faces tough questions about profitability before going public.

How OpenAI's Burn Rate Compares to Past Tech Giants

Is OpenAI's burn rate historically unprecedented, or is this just how transformative tech companies scale? The leaked financials invite comparison to earlier tech eras.

Amazon's early losses: a misleading comparison

Amazon lost money for years before becoming profitable, but its peak burn rate never exceeded $1.4 billion in a single year (adjusted for inflation). OpenAI is losing 15 times that. Amazon's capital intensity was also far lower — it sold books and general merchandise, not custom supercomputers costing billions to build and hundreds of millions annually to operate.

Uber: the closest parallel

Uber lost $8.5 billion in 2019, its worst year, before its IPO. OpenAI's $21 billion operating loss is more than 2.5 times that figure. Even adjusting for inflation and scale, OpenAI's burn is in a different league — and Uber had the advantage of a proven unit economics model (each ride eventually profitable) that OpenAI has not yet demonstrated.

FAQ: OpenAI Leaked Financials

How much money does OpenAI actually lose?

Based on the leaked financials, OpenAI lost approximately $21 billion on operations in 2025, with a total net loss of $39 billion when including one-time charges and non-cash items like stock-based compensation. The operating loss figure is the most relevant for assessing the company's financial health.

Is OpenAI profitable in 2026?

No. The leaked documents suggest OpenAI will continue losing money through 2026 and likely into 2027. The company is not expected to reach operating profitability until at least 2028 under current spending trajectories.

How much has OpenAI spent on AI development total?

Since its founding in 2015, OpenAI has cumulatively spent over $70 billion on AI development, including the $34 billion spent in 2025 alone. This makes it the most expensive AI research project in history by a wide margin.

When is OpenAI planning its IPO?

The leaked financials were reportedly prepared as part of confidential IPO roadshow materials. Analysts now expect the IPO to be delayed from late 2026 to at least mid-2027 as the company works to address investor concerns about its burn rate.

Conclusion: What the Leaked Numbers Tell Us About OpenAI

The leaked financial records reveal a company at a crossroads. The $34 billion in annual spending and $39 billion net loss show just how expensive it is to lead in frontier AI — and how far the company still is from sustainable operations. Compute costs alone consume more than the company's entire revenue, creating a structural deficit that no amount of subscription growth can quickly fix.

The coming months will determine whether OpenAI can cut costs without sacrificing its competitive edge, or whether the path to IPO requires accepting a lower valuation and tighter investor constraints. Either way, the leaked financials mark a turning point in how the industry thinks about AI's economic sustainability.

The race to build general intelligence has never been more expensive — and the finish line keeps moving.

What's your take on OpenAI's financial situation? Drop your experience in the comments — do you think the IPO will happen on schedule, and what would you change about OpenAI's spending strategy if you were in charge?