This figure if accurate or interpreted correctly—is far below earlier estimates of training costs for large foundational models in the West, which can exceed hundreds of millions.
The paper stated that training the model relied on a computing cluster of Nvidia chips (512 H800 GPUs) and that the model focuses on logical reasoning and inference capabilities. The announcement was an update to a previous paper the company released in January, but this time it included the financial detail for the first time—raising questions about how such cost efficiency was achieved.
DeepSeek’s announcement caused confusion among some global investors when the company first floated the claim, triggering sell-offs of technology stocks linked to AI, such as chipmakers. Western institutions and officials questioned the accuracy of the numbers or the benchmarks used for comparison, with some arguing that such figures may not reflect full costs (including R&D, infrastructure, and product marketing).
For context, OpenAI CEO Sam Altman had previously said that training foundational models cost “much more” than $100 million, a statement often cited when comparing with DeepSeek’s numbers. The stark difference raises questions about cost optimization, engineering methods, computing tools, or even how expenses are calculated.
What does this mean for the AI market?
If DeepSeek’s numbers are correct, it would imply that training hardware costs could drop dramatically when effective models can be built cheaply. That could change the rules of competition and lower barriers to entry for small and mid-sized companies. If, however, the numbers are less accurate or based on different standards, the impact may be more symbolic: the perception of falling costs could ignite a new race for innovation and competition in deploying advanced applications rivaling tech giants.
DeepSeek’s announcement is a strong signal that the global AI race is far from settled. Technology is advancing quickly, and the details of costs and methodologies remain open to scrutiny and change. Markets, investors, and regulators will likely approach such news with caution until the technical and financial picture becomes clearer.