DeepSeek, founded in July 2023 by Liang Wenfeng, has quickly established itself as a disruptive force in the Artificial Intelligence landscape. The company’s rapid ascent culminated in the January 2025 release of its DeepSeek-R1 model and accompanying chatbot, which swiftly became the most downloaded app on Apple’s U.S. App Store by January 27.
The company’s success has been attributed to its ability to develop high-performance AI models at a fraction of the cost of its Western competitors. DeepSeek claims to use only about 10% of the computing power required by similar models from companies like Meta, a feat that has positioned it as a serious challenger to established AI giants.
On March 1, 2025, DeepSeek released detailed operating data that forms the basis of its provocative profit margin claim. According to the company, over a 24-hour period from February 27-28, 2025, its AI models V3 and R1 processed 608 billion input tokens and 168 billion output tokens. Using its pricing structure, which charges $0.14 per million input tokens for cache hits, $0.55 per million for cache misses, and $2.19 per million output tokens, DeepSeek calculated a potential daily revenue of $562,027.
With daily operating costs of $87,072, primarily from renting Nvidia H800 GPUs at $2 per hour per chip, the company arrived at its headline-grabbing 545% cost-profit ratio. However, as TechCrunch reports, DeepSeek itself acknowledges that its actual income is “substantially lower” than this theoretical calculation suggests.
This admission has not quelled the controversy surrounding the claim. Industry observers have raised several critical questions about the validity and implications of DeepSeek’s profit margin theory:
Cost Considerations
Experts point out that the calculation only considers operational costs for inference, ignoring significant expenses such as research and development, model training, and infrastructure investments. These costs often represent the largest portion of expenses for AI companies and can substantially impact overall profitability.
Terminology Debate
Critics argue that the term “profit margin” may be misused in this context, as a true profit margin (the ratio of profit to revenue) cannot exceed 100%. They suggest that “cost-profit ratio” or “return on investment” might be more appropriate terms.
Real-World Application
The theoretical nature of the calculation has been questioned, as it assumes that all processed tokens are billed at the premium rate. This doesn’t reflect real-world usage patterns where customers may opt for lower-priced models or benefit from discounts.
Competitive Implications
DeepSeek’s claim of developing high-performance models at a fraction of the cost of its competitors has raised questions about the future of AI economics. Industry watchers are keenly observing how this might influence future AI development strategies, particularly in terms of cost efficiency and resource utilization.
Regulatory Concerns
DeepSeek’s focus on research and apparent lack of detailed commercialization plans have allowed it to avoid some of China’s strictest AI regulations. However, this success has led to government restrictions in some countries, with Taiwan banning its use in government agencies due to national security concerns.
Market Impact
The company’s aggressive pricing strategy and open-weight approach (which provides less freedom for modification than true open-source software) have already caused significant market disruption. U.S. tech stocks, particularly companies like Nvidia, experienced sharp declines following DeepSeek’s rise to prominence.
Future Profitability
The sustainability of DeepSeek’s business model and its ability to maintain such high margins in a competitive market remain open questions. The industry experts are debating the potential for actual profitability at this scale.
Despite the controversy, DeepSeek’s transparency in releasing this data has positioned it as a notable player in the global AI market. The company’s innovative use of technology, including advanced caching techniques and algorithmic optimizations, has demonstrated the potential for significant efficiency gains in AI model operation.
As the AI industry continues to evolve rapidly, DeepSeek’s claims and achievements serve as a catalyst for broader discussions about AI profitability, pricing models, and the future of AI economics. While the true implications of their 545% profit margin theory remain to be seen, it’s clear that DeepSeek has succeeded in challenging conventional wisdom about the costs and potential returns of operating advanced AI models.
