DeepSeek, a Chinese AI startup, shared key financial metrics on Saturday, providing one of the few insights into the economics of the AI industry.
The company announced via X that its V3 and R1 models hit a theoretical profit margin of 545% over 24 hours on the final day of February.
Deepseek clarified on GitHub that the information it shared was not conclusive. It stated that the actual profit is much lower than the theoretical margin suggests.
The AI firm stated that the difference is due to different factors, including the fact that only a small set of its services are monetized and that it offers discounts during off-peak hours.
Moreover, the cost estimates here don’t include the massive R&D and computational costs associated with training and tuning its proprietary AI models. So, while net income margins look very high in the numbers we’re seeing, they don’t reflect the broad reality of the economic life of a large-scale AI firm.
DeepSeek’s mouth-watering profits are largely theoretical
While the enticing profit margins that DeepSeek suggests are mostly theoretical, the firm’s disclosure arrives at a pivotal moment, with technology investors scrutinizing the financial sustainability of AI startups and their business models.
Investors are more focused than ever on whether AI firms can establish sustainable revenue streams amid high infrastructure costs, fierce competition, and evolving regulations.
The revelation from DeepSeek — with its caveats — adds to the discussion of a bigger question: Can AI companies transition from research-driven enterprises to profitable businesses? As the sector matures, such disclosures could provide valuable insight into the economics of AI as it evolves and the various monetization strategies that may prove viable in an industry where breakthroughs often involve a significant financial outlay.
Still, privacy and political concerns could soon temper the hype around DeepSeek. If the debacle around TikTok in the US is any indication of lawmakers’ attitudes toward Chinese apps, then DeepSeek may not be long for American markets.
AI firms have several revenue model trials
AI firms are using multiple revenue models for trials. From OpenAI Inc. to Anthropic PBC, AI companies are scrambling to try different revenue models to transform their innovative breakthroughs into viable businesses.
These approaches include subscription-based plans, pay-as-you-go pricing, and licensing fees for enterprise clients as they compete to create increasingly sophisticated AI products. However, the expensive cost of training, deploying, and maintaining these large-scale AI models has raised increasing investor concerns.
Question marks loom around the viability of these business models in producing consistently scalable profits, which has opened a broader discussion about whether AI companies can ever profit and/or become profitable in the short term.
One strategy DeepSeek outlined that enables them to keep costs down is optimizing computing power, including load-balance computing. This approach spreads traffic over servers and data centers to balance work and prevent bottlenecks, maximizing efficient resource utilization.
The company also highlighted its innovations that enable the models to process more data simultaneously, an improvement that can enhance the system’s overall performance. DeepSeek also highlighted its work on latency management, or the time between a user’s query and an AI’s response, which ensures user experiences go smoothly.
Cryptopolitan Academy: Want to grow your money in 2025? Learn how to do it with DeFi in our upcoming webclass. Save Your Spot
