Chinese AI startup DeepSeek recently declared that its AI models could be very profitable — with some asterisks.
In a post on X, DeepSeek boasted that its online services have a “cost profit margin” of 545%. However, that margin is calculated based on “theoretical income.”
It discussed these numbers in more detail at the end of a longer GitHub post outlining its approach to achieving “higher throughput and lower latency.” The company wrote that when it looks at usage of its V3 and R1 models during a 24-hour period, if that usage had all been billed using R1 pricing, DeepSeek would already have $562,027 in daily revenue.
Meanwhile, the cost of leasing the necessary GPUs (graphics processing units) would have been just $87,072.
The company admitted that its actual revenue is “substantially lower” for a variety of reasons, like nighttime discounts, lower pricing for V3, and the fact that “only a subset of services are monetized,” with web and app accessing remaining free.
Of course, if the app and website weren’t free, and if other discounts weren’t available, usage would presumably be much lower. So these calculations seem to be highly speculative — more a gesture towards potential future profit margins than a real snapshot of DeepSeek’s bottom line right now.
But the company is sharing these numbers amidst broader debates about AI’s cost and potential profitability. DeepSeek leapt into the spotlight in January, with a new model that supposedly matched OpenAI’s o1 on certain benchmarks, despite being developed at a much lower cost, and in the face of U.S. trade restrictions that prevent Chinese companies from accessing the most powerful chips. Tech stocks tumbled and analysts raised questions about AI spending.
DeepSeek’s tech didn’t just rattle Wall Street. Its app briefly displaced OpenAI’s ChatGPT at the top of Apple’s App Store — though it’s subsequently fallen off the general rankings and is currently ranked #6 in productivity, behind ChatGPT, Grok, and Google Gemini.