On Saturday, Chinese AI startup DeepSeek revealed cost and revenue estimates for its popular V3 and R1 models, showing a theoretical cost-profit ratio of up to 545% per day. This revelation further challenged the economics of U.S. AI companies like OpenAI that spend billions on cutting-edge chips.
What Happened: In a GitHub post, DeepSeek detailed that its AI inference costs-associated with running trained models-amount to $87,072 per day, assuming a rental price of $2 per hour for each Nvidia Corp.'s (NASDAQ: NVDA) H800 chip, reported Reuters.
In contrast, it estimates daily revenue of $562,027, equating to an annualized revenue of over $200 million.
However, DeepSeek acknowledged that "actual revenue is substantially lower" due to factors such as free web and app access, lower fees for off-peak usage, and the fact that some services remain unmonetized.
Why It's Important: The latest development follows earlier disclosures that DeepSeek spent under $6 million on chips to train its models-far less than U.S. rivals.
AI stocks outside China tumbled in January as investors reassessed the capital requirements for AI development. Nvidia's market value plunged by a record-breaking $593 billion in a single day-the largest loss ever for a Wall Street company.
DeepSeek's strategic pricing, including off-peak hours discounts last week, has also disrupted the AI industry.
Meanwhile, OpenAI CEO Sam Altman has also revealed that his company is incurring losses on its ChatGPT Pro plan, priced at $200 per month. At the time, it was also reported that despite raising $20 billion since its inception, OpenAI has yet to turn a profit.
Last year, it was reported that OpenAI, backed by Microsoft Corp. (NASDAQ: MSFT), intends to increase the price of ChatGPT over the next five years.