Most of DeepSeek’s revenue now comes from API access rather than its free consumer chatbot, according to the reporting. The company has continued to make its model weights publicly available while charging developers and businesses that use its hosted infrastructure.
DeepSeek recently raised API prices, with some models becoming 2.3 to 4.5 times more expensive. CEO Liang Wenfeng told investors that the increase did not reduce the company’s customer base, according to the report. Even after the adjustment, DeepSeek’s models remain among the lower-priced major language models.
The company’s newer V4.1-Flash and V4-Flash models were cited as part of that growth, with both designed to support agent-style workloads. The revenue figure suggests DeepSeek is converting interest in its models into sustained paid usage rather than relying solely on adoption of its free releases.
DeepSeek has also reported unusually high margins on its API operations. Previous reporting put gross margin for API access at 82.9% during the first seven months of the year. The company attributes part of that performance to infrastructure efficiency.
Its compute allocation also differs from what might be expected from a company generating most of its revenue through inference. More than 70% of DeepSeek’s computing resources are being used for model training, while less than 30% is allocated to inference, according to the reporting.
Liang has said smaller DeepSeek models can run effectively on gaming GPUs, potentially allowing more specialized hardware to remain available for training. He also expects the company to begin using Huawei chips for training as early as the fourth quarter of 2026.
DeepSeek was founded in 2023 by Liang, who also runs quantitative hedge fund High-Flyer. The company has built its reputation around relatively lean operations and a strategy that combines open model releases with paid hosted access.
The $1 billion figure remains an annualized run rate rather than audited yearly revenue, and DeepSeek has not publicly confirmed it. The reporting is based on people with knowledge of the company’s financials, while its next funding round has not yet closed.
Even with those caveats, the increase from less than $500 million to roughly $1 billion in annualized revenue shows how quickly DeepSeek’s business has expanded. The company is no longer simply an open-weight challenger attracting attention through model releases; its API operation is becoming a substantial commercial business in its own right.
This analysis is based on reporting from dealroom.co.
Image courtesy of Dado Ruvic.
This article was generated with AI assistance and reviewed for accuracy and quality.
