DeepSeek revenue surges 10-fold in 7 months

Summary

DeepSeek increased its revenue approximately tenfold year-over-year in the first seven months of 2026, reaching 475 million yuan. This growth was driven by the adoption of open-weight models and aggressive API pricing. However, the company reported a net loss of 715 million yuan and spent 11 billion yuan on AI infrastructure.

Why it matters

The article signals a structural shift in the large language model (LLM) market. Open-weight models have surpassed closed models in token usage on major developer platforms like Vercel. DeepSeek's cost-efficiency strategy is squeezing the profit margins of incumbent players such as Anthropic and Moonshot AI. If the company completes a 50 billion yuan capital raise at a 500 billion yuan valuation and files for an IPO on the STAR Market, it will solidify its position as a viable, capital-efficient alternative to the US-centric AI stack.

Highlights

• Open-weight models surpassed closed models for the first time, accounting for 62% of Vercel Gateway token usage.
• An 82.9% API gross profit margin demonstrates that inference cost optimization is translating into genuine profitability.
• AI infrastructure spending increased ninefold year-over-year to 11 billion yuan, while the net loss narrowed compared to 2025.

Critical take

The headline claiming a tenfold revenue increase is misleading. DeepSeek remains deeply unprofitable. Its 11 billion yuan infrastructure expenditure dwarfs its 475 million yuan revenue. This suggests that the 'efficiency' narrative relies on sustainable usage assumptions that cannot be validated by a single platform's (Vercel) snapshot of token share. Furthermore, the article relies on a single secondary source (The Information) for all financial data, with no mention of independent audits or company confirmation.