Epoch AI’s latest research indicates that profits generated by one AI model are “outweighed” by the significant costs of creating its successor.
A recent study from the non-profit research institute Epoch AI casts significant doubt on the long-term profitability prospects of AI companies, particularly OpenAI.
Authored by Jaime Sevilla, Hannah Petrovic, and Anson Ho, the research paper posits that even if an AI model generates sufficient revenue to cover its operational and R&D expenses, these gains are typically overshadowed by the investment required for the subsequent model’s development. Consequently, the study concludes that “companies can lose money each year” despite individual models generating revenue.
The study aims to address key questions: What is the profitability of operating AI models? Do models sustain profitability throughout their entire lifecycle? And will AI models ultimately achieve profitability?
For the first question, researchers devised a case study referred to as the GPT-5 bundle. This bundle encompassed all OpenAI products available during GPT-5’s period as the leading model, such as GPT-5, GPT-5.1, GPT-4o, ChatGPT, and its API. They then estimated the associated revenues and operational costs. The data used for these figures was compiled from various sources, including statements from OpenAI and its personnel, as well as reports from major media outlets like The Information, CNBC, and the Wall Street Journal.
According to the researchers, the revenue projection was “relatively straightforward”. Given that the bundle covered all of OpenAI’s models, it represented the company’s cumulative revenue during GPT-5’s active period from August to December of the previous year, totaling $6.1 billion.
However, they noted that “at first glance, $6.1 billion sounds healthy, until you juxtapose it with the costs of running the GPT-5 bundle.” The report identified four primary cost categories, with inference compute being the largest at $3.2 billion. This figure relies on publicly available estimates of OpenAI’s total inference compute expenditure for 2025, assuming compute allocation during GPT-5’s operational period was proportionate to the revenue generated within that timeframe.
Additional expenses included staff remuneration ($1.2 billion), sales and marketing activities ($2.2 billion), and legal, office, and administrative overheads totaling $0.2 billion.
Understanding the Financial Calculus
Regarding methodologies for profit assessment, the paper highlighted that “one option is to look at gross profits.” This approach solely accounts for the direct operational costs of a model, which, in this scenario, amounted to $3.2 billion for inference compute. With a revenue of $6.1 billion, this yields a $2.9 billion profit, or a 48% gross profit margin, consistent with other projections. While less than typical software companies, this margin is deemed sufficient for long-term business development.
Essentially, the authors concluded that “running AI models is likely profitable in the sense of having decent gross margins.”
Nonetheless, this perspective doesn’t reveal the complete picture.
The paper asserted that while considering only gross margins, “on those terms, it was profitable to run the GPT-5 bundle.” However, it questioned if this was enough to recover development expenses. Theoretically, this could be achieved by continuous operation until sufficient revenue is generated. Yet, practically, models often have brief lifespans, limiting revenue accumulation, particularly if they are superseded by competitors’ offerings, necessitating replacement.
The authors explained that the key lies in juxtaposing gross profits with the company’s R&D investments, which amounted to nearly $3 billion. They stated, “To evaluate AI products, we need to look at both profit margins in inference as well as the time it takes for users to migrate to something better. In the case of the GPT-5 bundle, we find that it’s decidedly unprofitable over its full lifecycle, even from a gross margin perspective.”
Addressing the overarching question of AI model profitability, the paper emphasized that “the most crucial point is that these model lifecycle losses aren’t necessarily cause for alarm. AI models don’t need to be profitable today, as long as companies can convince investors that they will be in the future. That’s standard for fast-growing tech companies.”
The authors ultimately concluded that profitability remains highly achievable because “compute margins are falling, enterprise deals are stickier, and models can stay relevant longer than the GPT-5 cycle suggests.”
When queried about the market’s willingness to sustain irrationality until OpenAI reaches solvency, Jason Andersen, VP and principal analyst at Moor Insights & Strategy, responded, “it’s possible, but there is no guarantee. I believe in 2026 you will see refinements in strategy from these firms. In my brain, there are three levers that OpenAI and other general-purpose AIs can use to improve their financial position (or at least slow the burn).”
He identified pacing as the first lever, stating, “and I think that is happening already. We saw major model drops at a slower pace last year. So, by slowing down a bit, they can reduce some of their costs or at the very least spread them out better. Frankly, customers need to catch up anyway, so they can plausibly slow down, so the market can catch up to what they already have.”
Andersen pointed to diversifying their product portfolio as the second lever, with the third being the generation of revenue from other software providers.
Regarding whether OpenAI and its counterparts can persist until AI reaches true effectiveness, he commented, “OpenAI and Anthropic have the best chance of going long and staying independent. But, that said, I also want to be cautious about what ‘truly effective’ means. If you mean truly effective means achieving AGI, it’s theoretical, so probably not without major breakthroughs in hardware and energy. But if ‘effective’ means reaching profitability over a period of years, then yes, those two have a shot.”
He elaborated that the challenge to achieving profitability “will be finding a way to compete and win against companies that have welded their future to AI. Notably, Google, Microsoft, and X have now made their models inextricable to their other products and offerings. So, is there enough time and diversification opportunities to compete with them? My guess is that a couple pure plays will do well and maybe even disrupt the market, but many others won’t make it.”
Scott Bickley, an advisory fellow at Info-Tech Research Group, characterized the paper’s conclusions as “very linear” and reliant on short-term analysis. He noted that OpenAI has been “pretty open about the fact they are not profitable currently. What they pivot to is this staggering chart of how revenues are going to grow exponentially over the next three plus years, and that’s why they are trying to raise $200 billion now to build up infrastructure that’s going to support hundreds of billions of dollars of business a year.”
Interconnected Fates with OpenAI
Bickley estimated that OpenAI’s total financial obligations, arising from partnerships with Nvidia, hyperscale providers, and data center expansions, now amount to $1.4 trillion. He remarked, “They’re trying to make themselves too big to fail, to buy the long runway they’re going to need for these investments to hopefully pay off over the course of years, or even decades.”
Currently, he explained, the company is “shoring up the balance sheet. They’re trying to build everything they can to buy runway ahead. But either they wildly succeed beyond any of our imagination, and they come up with applications that I can’t envision are realistic today, or they fail miserably, and they’re guaranteed that everyone can buy a chunk of the empire for pennies on the dollar or something to that effect. But I think it’s either boom or bust. I don’t see a middle road.”
Bickley observed that, at present, all significant vendors have “tied their fortunes to OpenAI, which is exactly what Sam Altman wanted to have happen. He’s going to force the biggest players in the space to help him be successful.”
Should the company ultimately fail, he foresees minimal impact on businesses utilizing AI initiatives developed by OpenAI. He stated, “Regardless of what happens to the commercial entity of OpenAI, the intellectual property that’s been developed, the models that are there, are going to be there. They’ll fall under someone’s control and continue to be used. They’re not in any danger of not being available.”
