• OpenAI's compute margin has climbed to around 70% in October 2025, up from roughly 52% at the end of 2024 and double its January 2024 level.
  • The improvement reflects falling compute costs per request and a shift toward higher-margin paid offerings and enterprise contracts.
  • Despite these gains, the company remains unprofitable overall, with aggressive spending on infrastructure and R&D continuing amid intense competition.

OpenAI has significantly improved the profitability of its core AI services, according to internal financials reported by The Information. The company's compute margin—which measures revenue after model-running costs—reached about 70% in October, up from approximately 52% at the end of 2024 and roughly double the level seen in early 2024. People familiar with the matter attribute the gains to a combination of technical optimizations that have lowered per-token inference costs and a deliberate mix shift toward higher-priced paid and enterprise products.

"We're seeing a clear move away from the free tier toward more sustainable revenue streams," said one source close to the company, who requested anonymity because the figures are not public. OpenAI declined to comment on the specific margin numbers, but the trend aligns with broader reporting that the firm is aggressively trying to improve unit economics while still burning cash at an extraordinary rate.

Efforts to restructure its cost base have hit a snag in some areas, however. Despite leading the AI boom and commanding a valuation above $500 billion, OpenAI remains deeply unprofitable on a net basis. Financial documents reviewed by The Wall Street Journal and summarized by Fortune indicate the company is on track for a net loss of roughly $9 billion in 2025, with cumulative cash burn potentially reaching $115 billion by 2029 before turning cash-flow positive. Without a sustained improvement in both margins and revenue growth, the company would face mounting pressure from investors and competitors alike.

Industry observers note that the margin improvement comes as OpenAI prioritizes enterprise customers and explores new funding and infrastructure partnerships. The company has told CNBC it reached $10 billion in annual recurring revenue by mid-2025, with third-party analysis suggesting that figure may have climbed to $12 billion or more by July. Yet this breakneck growth has required massive spending on data centers, chips, and talent, keeping overall profitability out of reach for now.

Competitive dynamics are intensifying. Google (GOOGL) and Anthropic are both pushing hard on their own AI offerings, with Anthropic targeting breakeven around 2026–2027 on a more conservative burn path. This has forced OpenAI to balance margin expansion with continued investment in next-generation models and products, including Sora 2, the Atlas browser, and explorations into consumer hardware and robotics. "You can't afford to stand still in this market," said an analyst who follows the sector. "Every efficiency gain gets plowed back into the race."

Regulatory scrutiny adds another layer of complexity. Governments in the U.S., EU, and UK are drafting AI safety and competition rules that could affect product design and deployment. At the same time, OpenAI's infrastructure build-out intersects with industrial policy and semiconductor strategy, drawing attention from antitrust regulators as the firm deepens partnerships with major cloud and chip players.

Looking ahead, internal projections suggest OpenAI is aiming for revenue to soar from around $10–13 billion in 2025 to $100 billion or more by 2028, with cash-flow positivity targeted for 2029–2030. Whether demand can keep pace with such aggressive spending remains an open question. For now, the improved compute margin offers a glimmer of operational efficiency in a landscape still defined by vast ambition and even vaster costs.

Correction: An earlier version of this article misstated the timeline for OpenAI's margin improvements. The 70% compute margin was reported for October 2025, not late 2024.