Who Really Profits from AI Developments
172A new report from Epoch AI, a non-profit research group, challenges the idea that AI companies like OpenAI will become profitable in the traditional sense. The study suggests that while running AI models can generate significant revenue, the costs to develop the next big model often outweigh those earnings. This means companies might make money on individual models but still lose money overall each year.
Examining AI Profitability Through a Case Study
The researchers created a hypothetical scenario called the GPT-5 bundle, which includes all of OpenAI’s offerings available during GPT-5’s lifetime. This bundle covers GPT-5, GPT-5.1, GPT-4o, ChatGPT, and the API. They used publicly available data, including statements from OpenAI and media reports, to estimate the revenue and expenses associated with running this bundle.
The total revenue from the bundle during GPT-5’s four-month period, from August to December last year, was estimated at $6.1 billion. While that sounds impressive at first, the report emphasizes that these figures must be weighed against the costs involved in maintaining and developing the models. The main expenses include inference compute costs, staff salaries, marketing, and administrative expenses.
Breaking Down the Costs and Profits
The biggest expense was inference compute, estimated at around $3.2 billion. This cost reflects the processing power needed to run the models and was calculated based on public estimates of OpenAI’s spend in 2025. Other costs included $1.2 billion for staff compensation, $2.2 billion for sales and marketing, and $200 million for legal and administrative expenses.
When looking at gross profit, the researchers focused only on the direct costs of running the models—mainly the inference compute. Subtracting this from revenue gives a gross profit of about $2.9 billion, which translates to a margin of roughly 48%. This suggests that operating AI models can be profitable in terms of gross margins, similar to other software businesses.
Profitability Is More Complex Than It Seems
However, the report argues that gross margins don’t tell the whole story. Just because a model makes money on paper doesn’t mean the company is actually profitable overall. The costs of developing and improving the models—such as research, infrastructure, and future upgrades—are not included in these calculations. These ongoing investments can quickly eat into any short-term profits.
The authors point out that, in theory, companies could keep running models that break even or make a little profit, but that doesn’t necessarily mean they’re recouping the huge investments needed to develop the next generation. Over time, these costs can outweigh the earnings, making true profitability difficult to achieve in the long run.
Overall, the study suggests that while AI models can generate substantial revenue, the high costs associated with building and maintaining them mean that AI firms like OpenAI might not be profitable in the traditional sense. Instead, they might be investing heavily in future technologies, with current earnings serving more as a way to fund ongoing development rather than create immediate profits.
What do you think?
It is nice to know your opinion. Leave a comment.