OpenAI Makes $25B a Year and Loses $14B — Inside the Math (2026)

OpenAI runs at roughly $25 billion in annualized revenue — about $2B per month, essentially flat since February 2026 — while leaked audited financials showed a $20.9B operating loss on $13.07B of booked 2025 revenue, a confidential S-1 filed June 8, 2026, and a rival that just passed it on the top line.

Growing revenue ~7x in two years is no longer the hard part for OpenAI. The hard parts are that the growth curve went flat this spring, the losses are still bigger than the business, and Anthropic's run rate crossed OpenAI's in April 2026 for the first time.

OpenAI Revenue 2026: The $25B ARR Breakdown by Product

OpenAI's ~$25B annualized run rate in mid-2026 breaks down into roughly $17B from ChatGPT subscriptions, $6.5B from API consumption, and $1.5B from Sora, the new ads pilot, and licensing. ChatGPT Plus at $20/month is still the largest single product line, but Enterprise and Team seats grew the fastest — and enterprise revenue now exceeds 40% of the total, on pace to reach parity with consumer revenue by the end of 2026.

Sources: OpenAI investor disclosures, The Information, FT, Bloomberg, Sacra, reported leaks Q1–Q2 2026. 2025 and 2026 figures are annualized run rates, not GAAP recognized revenue; *2024 is full-year booked revenue. Booked 2025 revenue was $13.07B per leaked audited financials.

From $1B to $25B — and Then a Flat Line

The clearest way to see what happened: OpenAI passed $1B annualized revenue in mid-2023, booked $3.7B in 2024 and $13.07B in 2025, exited 2025 at a $21.4B run rate, and hit ~$25B by February 2026. Then the curve went flat. After adding roughly $15B of new run rate in the eight months to February, OpenAI has added close to zero since — a five-plus-month plateau that is the single most important change in its 2026 story. For context on how growth rates translate to AI vs SaaS valuation multiples, see the full breakdown.

Why the stall? ChatGPT has largely saturated its addressable consumer base — 1 billion monthly actives as of June 2026, roughly 900 million weekly — so there are only so many net-new subscribers left to add, and enterprise deals close on procurement timelines, not subscription toggles. The $200/month Pro tier (500K+ subscribers, a $1.2B+ ARR line on its own) and enterprise seats past 7 million keep the base compounding slowly, but the hypergrowth phase of consumer conversion is over. Track broader adoption trends on the Enterprise AI Adoption dashboard.

The Leaked Financials: $13.07B of Revenue, a $20.9B Operating Loss

In June 2026, OpenAI's audited financial statements leaked — first reported by Ed Zitron and verified by the Financial Times and Fortune. The headline: a $20.92B operating loss on $13.07B of booked 2025 revenue — a loss larger than the revenue that produced it. Total costs came to roughly $34B, split between $19.18B of R&D and $5.73B of sales and marketing, with $17.2B flowing to Microsoft for Azure compute — more than OpenAI's entire annual revenue, to a single vendor.

The net loss was larger still: $38.5B, once a one-time, non-cash charge of roughly $41.55B tied to OpenAI's 2025 nonprofit-to-for-profit conversion is layered in. Strip out that structural noise and the $20.9B operating loss is the number that matters for anyone modeling the business. It also reframes the run-rate language OpenAI prefers: a $25B annualized pace is a projection off the latest month; $13.07B is what was actually booked and audited in 2025.

OpenAI Burn Rate and the Path to Profitability

OpenAI is on pace for roughly $14B in 2026 losses on ~$25B of run-rate revenue, after burning $3.7B in Q1 2026 alone against $5.7B of quarterly revenue. Internal projections surfaced to the FT and The Information show cumulative losses of roughly $115B through 2029 before the company turns cash-flow positive around 2029–2030 at $125B+ annual revenue. For context, Amazon burned roughly $3B cumulative in its first decade; Uber burned about $25B before GAAP profitability. OpenAI is in a different category entirely — and analysts including HSBC have publicly questioned whether the 2029–2030 timeline holds, citing a funding gap between committed infrastructure spend and disclosed capital raised.

Three numbers explain the bulk of the burn. OpenAI's 2026 compute spend with Microsoft Azure is roughly $13B (after $17.2B in 2025). The Stargate JV with SoftBank, Oracle, and MGX is in the early stages of a $500B multi-year buildout. Talent costs run roughly $4B annually across ~4,500 employees — about $900K fully loaded per head. Track this against the broader AI infrastructure spend on our AI Spending Dashboard.

OpenAI vs Anthropic, Google, and xAI: Who's Winning Revenue in 2026?

This is the biggest change since this post was first published: OpenAI is no longer the run-rate leader. Anthropic passed OpenAI in April 2026 at roughly $30B versus $25B, and reached $47B by May 29, 2026 per its Series H announcement — nearly double OpenAI's figure, built on an 80%+ enterprise and API mix versus OpenAI's consumer-heavy base. Full breakdown in our piece on how Anthropic hit a $47B run rate.

*Google reports AI revenue inside Cloud and Workspace and doesn't break out Gemini-specific ARR. Anthropic figures per its Series H announcement (May 29, 2026); OpenAI figures per The Information, Sacra, and Epoch AI tracking. Both companies' ARR figures are annualized run rates, not booked revenue — on trailing booked 2025 revenue, OpenAI ($13.07B) was still larger than Anthropic (~$9B).

The $100M Ads Bet: OpenAI's Third Revenue Leg

The most interesting new line in OpenAI's 2026 mix isn't enterprise — it's ads. The ChatGPT ads pilot, running only in the free and Go tiers, surpassed $100M in annualized revenue in under six weeks with more than 600 advertisers, even though fewer than 20% of eligible free-tier users saw an ad daily during the test window. It's a rounding error next to $25B in ARR and a $13B compute bill, but it's the fastest-scaling new revenue line OpenAI has launched since ChatGPT Plus — and with roughly 950 million free users generating near-zero direct revenue today, it's the clearest lever left if subscription growth stays flat.

OpenAI Revenue Per User and Unit Economics

With roughly 900 million weekly active users and ~$25B annualized revenue, OpenAI's blended revenue per weekly active user is about $28/year — but the distribution is wildly bimodal. The vast majority of users are free and generate almost nothing directly (the ad pilot is just starting to change that). The 50M+ paid subscribers average roughly $500/year — heavily skewed by Enterprise contracts and the $200/month Pro tier.

• Plus tier ARPU: $240/year ($20/mo × 12) — the volume base of the paid mix
• Pro tier ARPU: $2,400/year ($200/mo × 12) — 500K+ subscribers — a $1.2B+ run-rate line on its own
• Enterprise: 7M+ workplace seats deployed; enterprise revenue now >40% of total and heading toward parity with consumer
• API: highly variable; top customers spend tens of millions per year each

Gross margin improved to roughly 39% in Q1 2026, up from 33% a year earlier, because the inference cost per query has dropped roughly 95% since GPT-4's launch in early 2023. The problem is that the improvement is being swamped by compute and talent spending growing as fast as revenue — which is why the operating loss keeps widening even as unit economics get better.

What OpenAI's Revenue Numbers Mean for Investors

At an $852B valuation against ~$25B ARR, OpenAI is priced around 34x run-rate revenue — well above Anthropic's roughly 21x despite the slower growth, and a multiple that only makes sense if the 2029–2030 profitability story arrives on schedule. The bull case: OpenAI hits its plan — $125B revenue by 2029, steady-state operating margins in the 30–40% range, and the S-1 filed June 8, 2026 converts into a 2027 listing at or above the private mark. The bear case: compute costs scale roughly with revenue, the consumer base is saturated, and a flat top line meets an accelerating cost base. Compare the broader private AI revenue rankings.

For LPs in venture funds with OpenAI exposure (Sequoia, Thrive, Founders Fund, a16z, Khosla), the $852B mark is a meaningful chunk of net TVPI on 2018–2021 vintage funds — real but unrealized, with the path to DPI running through either the 2027 IPO window or continued secondary tenders at or above current marks. Track the frontier AI valuation race on our AI Valuations dashboard.