Chinese AI startup DeepSeek has officially announced an upcoming "significant" API cost increase. Specific tariff plans will be announced later, but the company has already urged customers to adjust the use of the services in advance. This decision was made against the background of a large-scale financing round and active preparations for a possible listing on the stock exchange this year. After the completion of the June round, the company's founder, Liang Wenfeng, became the richest entrepreneur in the world among the creators of AI models. Previously, a startup relied on the cheapest possible solutions, but now investors expect a transition to a more profitable business model.

The current pricing policy has made DeepSeek the main factor in the price war in the generative AI market. The V4 Flash model costs only 14 and 28 cents per million tokens. For comparison, the Moonshot Kimi K3 costs $3 and $15, while the Anthropic Fable 5 costs $10 and $50. Ultra-low tariffs have given rise to the term "DeepSeek death zone," in which more expensive models lose their competitiveness. Three weeks ago, the company introduced time-of-day tariffs to balance the load, but the current move means a complete rethink of positioning. Additionally, DeepSeek plans to build a large data center with a capacity of 1 GW in Inner Mongolia.

The market took the news ambiguously. Since all V4 Flash models are available under an open license, other providers will be able to host it on their own servers. Developer Michael Guo questioned the wisdom of the move, noting that the American models Meta Muse Spark and OpenAI GPT-5.6 Luna already offer similar or even lower prices. Bloomberg Intelligence analyst Robert Lee, in turn, drew attention to the continued high demand in China. DeepSeek and ByteDance continue to dominate the domestic market, increasing pressure on smaller competitors. We remind you that in May, DeepSeek already staged a price war, reducing tariffs for the flagship V4-Pro model by 75%.