In less than two years, the "profit margin" doubled from 35% to 70%. Can OpenAI actually be very profitable?

In less than two years, the "profit margin" doubled from 35% to 70%. Can OpenAI actually be very profitable?

As the AI industry's money-burning competition grows ever fiercer, OpenAI has demonstrated a remarkable improvement in operational efficiency. According to The Information's report on Sunday, the company's compute profit margin has soared from around 35% to 70% in less than two years, showing significant progress in controlling AI model operating costs.

This improvement comes at a critical time as OpenAI seeks up to $100 billion in financing. The substantial increase in compute profit margin means the company is generating more revenue per dollar spent on servers, a positive signal for potential investors.

CEO Sam Altman said candidly in a recent podcast interview, "If we didn’t continue to massively increase training costs, we’d have been profitable long ago." This statement highlights the core dilemma faced by the AI industry—the tough balance between pursuing technological breakthroughs and financial sustainability.

A similar trend of efficiency improvements is seen across the AI sector, with competitors like Anthropic also significantly optimizing their compute cost structures, reflecting a collective industry effort toward operational optimization.

Significant Improvement in Operational Efficiency, but Still Below Traditional Software Companies

OpenAI’s trajectory of compute profit margin improvement is impressive. According to insiders, this metric rose from about 35% in January 2024 to 52% by the end of last year, reaching around 70% in October this year. This profit margin reflects the portion of income retained after deducting the cost of running AI models for paid users.

Despite notable progress, the 70% compute profit margin is still far below the equivalent metric for publicly traded software companies. Traditional software firms can serve additional users, including free users, at very low cost, which points to further room for optimization for OpenAI.

The boost in efficiency mainly comes from three factors: falling compute leasing costs during the year, technical optimization of AI model operational efficiency, and revenue growth brought by higher-priced subscription tiers. After DeepSeek launched lower-cost AI models this February, OpenAI internally declared "red alert" status, prioritizing reducing server costs.

Industry Commonality: Anthropic Also Facing Cost Challenges

Pressure from compute costs is not unique to OpenAI. According to The Information’s analysis of Anthropic’s financial data, the company’s compute profit margin last year was about -90%, meaning operating costs far exceeded revenue.

However, Anthropic expects to improve this metric to about 53% by the end of this year, and in the most optimistic forecasts, next year’s compute profit margin could reach 68%. This trajectory suggests the entire AI industry is undergoing a similar process, starting from high costs and gradually optimizing.

In terms of overall server efficiency, Anthropic’s forecasts suggest it will surpass OpenAI. This is mainly because OpenAI must bear costs for hundreds of millions of non-paying chatbot users, while Anthropic’s free user base is much smaller. OpenAI must monetize these free users through ads or affiliate shopping commissions to make up the efficiency gap.

Stark Disparity in Compute Investments; Competitive Landscape Differentiates

The two companies differ drastically in scale of compute investment. Anthropic is projected to spend up to $60 billion between 2025 and 2028 on total compute costs, including developing new AI. This estimate does not include recent server leasing deals with Google and Microsoft.

By contrast, OpenAI is expected to spend $220 billion on servers during the same period, nearly four times Anthropic’s estimate. This massive investment reflects core leadership beliefs at OpenAI: server shortages are the biggest obstacle to the company’s growth and its pursuit of artificial general intelligence.

Altman emphasized in the podcast, "We cannot do this without compute. We are so acutely compute-constrained. If we had double the compute, I think we could have double the revenue right now."

Technological Weakness and Profitability Pressure Coexist

OpenAI also faces technical cost pressure from Google. Google uses custom tensor processing unit (TPU) chips to reduce costs, while OpenAI relies on Nvidia server chips, which are expensive. As previously reported by The Information, OpenAI leadership believes Google’s AI runs more efficiently, meaning Google faces less pressure to monetize non-paying users.

Faced with skepticism over its $1.4 trillion spending commitment versus $20 billion in income, Altman admitted that training cost growth still outpaces revenue growth. But he insists the company’s ongoing "compute deficit" status is precisely proof of strong demand.

Reportedly, OpenAI may face losses of about $120 billion before breaking even in 2028 or 2029. Altman affirms the company’s strategic focus is on expanding compute capacity supported by revenue growth, rather than cutting investments due to short-term losses.

Risk Notice and DisclaimerThe market carries risks, and investment must be cautious. This article does not constitute personal investment advice, nor does it address individual users’ specific investment objectives, financial situation, or needs. Users should consider whether any opinions, views, or conclusions in this article are appropriate for their particular circumstances. Investment is at your own risk.