Portuguese 
English 
Spanish 
2 min of reading 
0 comments 
A New Study Revealed That DeepSeek’s AI Training Did Not Cost US$ 6 Million, But Rather A Much Higher Amount. Discover!
A recent analysis by SemiAnalysis revealed that the actual cost of training DeepSeek’s artificial intelligence (AI) is significantly higher than previously thought.
While initial estimates suggested an investment of approximately US$ 6 million, the report indicates that the actual value reaches an impressive US$ 1.3 billion.
Unraveling The Myth Of US$ 6 Million
The initial estimate of US$ 6 million only considered pre-training expenses with GPUs, neglecting substantial investments in research and development, infrastructure, and other essential costs accumulated by the company.
-
Powering ChatGPT and the world’s data centers in 2026 will require more than 1,000 TWh of electricity — equivalent to Japan’s entire consumption.
-
Google bought 115 MW and six banks injected US$ 421 million: why AI data centers are choosing geothermal energy instead of solar and wind
-
OpenAI is preparing the launch of GPT-5.5-Cyber, a new Artificial Intelligence specialized in cybersecurity with restricted access to trusted selected specialists and institutions.
-
Federal government institutes official artificial intelligence policy with mandatory rules for ethical, transparent, and secure use throughout the public administration.
The report highlights that the total capital expenditure (CapEx) of DeepSeek on servers reaches approximately US$ 1.6 billion, with a considerable portion of this amount directed toward the operation and maintenance of its extensive GPU clusters.
Robust Infrastructure And Hardware Investments
DeepSeek has access to about 50,000 Hopper Series GPUs, including models such as H800s, H100s, and H20s, specific to each country, produced by NVIDIA in response to U.S. export restrictions.
This diversification in hardware inventory reflects the company’s strategic sourcing decisions and operational efficiency.
Organizational Structure And Operational Efficiency
Unlike some of the largest AI labs, DeepSeek operates its own data centers and adopts a streamlined model that contributes to its agility and efficiency. This capacity for rapid adaptation is vital in an increasingly competitive AI landscape.
In terms of performance, DeepSeek’s R1 model demonstrates reasoning capabilities comparable to OpenAI’s o1.
However, it is not considered the undisputed leader across all performance metrics. While DeepSeek’s pricing strategy has received praise, it’s important to note that Google’s Gemini Flash 2.0, with similar capabilities, proves to be even more cost-effective when accessed through API services.
This places DeepSeek in the challenging position of balancing performance and cost to ensure its future success.
A notable innovation highlighted in the report is the Multi-Head Latent Attention (MLA) technology, which significantly reduces inference costs by an impressive 93.3% through reduced use of key-value (KV) caching. This approach represents a major breakthrough toward more economical AI solutions.
Experts suggest that DeepSeek’s innovations are likely to be quickly adopted by Western AI labs seeking to remain competitive.
The Future Of Chinese AI
While there is optimism regarding potential improvements and gains in efficiency, SemiAnalysis warns of external challenges.
The report speculates that operating costs may decrease further, driven by DeepSeek’s ability to adapt quickly compared to its larger and more bureaucratic counterparts.
However, scaling operations amid rising U.S. export controls poses a significant hurdle that DeepSeek must cautiously overcome.
With information from SemiAnalysis.
Be the first to react!