DeepSeek’s ‘Theoretical’ Profit Margins Are Just That

The Chinese artificial intelligence startup that rocked global markets earlier this year with its low-cost and high-performance AI models has outlined a potential path to major profitability. The transparency is laudable — even if the operating word is “potential.”

Over the weekend, DeepSeek shared an eye-popping “theoretical” cost-profit margin of 545%. The revelation came as the closing update in the company’s weeklong show that gave the world an exceedingly rare look under the hood of an AI firm. On Saturday, it published a blogpost outlining its potential profit margins when looking at a 24-hour period of inferencing costs (essentially, the computing power and related real-time operating expenses) compared to user requests for its two latest models, V3 and R1.

DeepSeek claimed that if all of these sales were billed at R1’s pricing, it could achieve a significant “theoretical” revenue while still charging significantly less than competitors. The figures come with some major caveats, which the company acknowledges. “Actual revenue is substantially lower,” because only a smaller subset of services are monetized (its app and web version remain free), it offers significant off-peak discounts, and its pricing for the V3 model is lower than for the R1. It also doesn’t seem to account for other costs, including research and development or training. In other words, this all makes its calculations highly “theoretical.”

Still, it comes at a time when global investors remain skeptical about how Silicon Valley’s sky-high spending on AI will eventually pay off. DeepSeek’s granular level of transparency stands in stark contrast to its US peers. It offers perhaps a glimmer of hope that generative AI has a path to profitability. But breaking down DeepSeek’s report reveals some existential questions about whether this can be achievable any time soon from China to Silicon Valley.

For starters, the cheap cost of deploying DeepSeek’s models has been a key driver of widespread adoption. If the company did charge all of its user requests at the higher R1 pricing, it would likely be receiving significantly fewer of them.

Competition is mounting in the global AI industry, but the pressure is especially fierce in China’s hyper-competitive market. The domestic race to woo consumers has increasingly compelled companies to offer their services for free. Last month, tech giant Baidu Inc. announced it was making its Ernie AI chatbot free to use in a bid to regain momentum. The ubiquity of no-cost AI tools in China is driving companies to slash prices while also providing little incentive for consumers to cough up money for paid services or subscriptions.