Who Really Benefits from AI Profits
182A new study from Epoch AI, a non-profit research group, challenges the idea that AI companies like OpenAI will eventually turn a profit. The researchers suggest that while running an AI model can cover some costs, the ongoing expenses of developing new models often outweigh the income. This raises questions about whether these companies are truly profitable in the long run or just breaking even on their existing products.
Understanding the Profitability of AI Models
The study focused on a hypothetical scenario called the GPT-5 bundle, which included all of OpenAI’s products during GPT-5’s release period. This included models like GPT-5, GPT-4o, ChatGPT, and the API service. The researchers estimated the total revenue from these offerings over a few months, based on publicly available data and reports from media outlets.
The total revenue for the GPT-5 bundle was calculated at around $6.1 billion, covering the period from August to December last year. While that sounds impressive, the real question is whether this revenue was enough to cover the costs of running and developing these models. The study broke down the main expenses and found that the costs were significant, especially for computational resources, staff salaries, and marketing efforts.
Costs and Margins in AI Operations
The biggest expense was inference compute, the power needed to run the models, which alone cost about $3.2 billion. This estimate was based on public data about OpenAI’s compute spending and assumed that costs were proportional to the revenue generated during GPT-5’s lifetime. Other costs included staff compensation at $1.2 billion, sales and marketing at $2.2 billion, and administrative expenses of $200 million.
When looking at gross profit—meaning revenue minus the direct costs of inference compute—the numbers look promising. The study found a gross profit of around $2.9 billion, which translates to a margin of about 48%. This suggests that, at least from a simple revenue versus direct costs perspective, running these models can be profitable. However, this doesn’t tell the full story about overall profitability.
The researchers pointed out that gross margins only consider direct expenses and ignore the heavy costs involved in developing future models. So, while the existing models might generate decent margins, the investment needed to create the next generation can wipe out those gains. Essentially, companies might be making money on each model, but losing money overall when factoring in ongoing development costs.
The paper emphasizes that profitability in AI isn’t just about current revenues and costs. It depends heavily on whether companies can recoup their huge investments in research, hardware, and talent. The findings suggest that, despite some revenue being generated, AI firms like OpenAI might not be profitable in the traditional sense, especially when considering long-term development expenses.
In conclusion, the study challenges the common belief that AI companies are on the path to big profits. Instead, it indicates that while they can make money on individual models, the overall financial picture is more complicated. The real winners may not be the AI firms themselves, but the broader ecosystem that supports their research and development efforts.
What do you think?
It is nice to know your opinion. Leave a comment.