China’s DeepSeek Warns of ‘Significant’ Price Increase After Undercutting US AI Rivals(Yicai) Aug. 6 -- DeepSeek, the Chinese startup that shot to global fame for developing high-performance open-weight artificial intelligence models at a fraction of the cost of US competitors, has warned users of a “significant” price increase in the near future.
The Hangzhou-based company, which did not disclose the size of the price hike for application programming interface services in its announcement today, advised users to plan their usage accordingly, adding that pricing details will be released separately.
Other Chinese AI developers have also increased API fees this year, reflecting broader cost pressures across the domestic AI industry. Tencent Cloud has raised them twice, and Zhipu AI three times.
DeepSeek's move follows the introduction of peak and off-peak pricing in mid-July for its V4 series, launched in April, which includes the high-performance V4 Pro and lower-cost V4 Flash models, and comes as demand for computing resources continues to outpace supply.
Under that pricing mechanism, API prices during daily peak hours -- from 9 a.m. to noon and from 2 p.m. to 6 p.m. -- are twice the off-peak rate. The standard price for V4 Pro is 87 US cents per million output tokens, while V4 Flash costs 28 US cents per million output tokens.
Goldman Sachs estimated at the time that the new pricing structure would result in an average blended price of about 35 US cents per million tokens for V4 Pro and 12 US cents for V4 Flash. The market largely took it as a sign that DeepSeek was improving operational efficiency and managing demand more exactly, rather than abandoning its low-cost strategy.
The planned price increase may also signal the upcoming release of the official version or broader release of V4-Pro, which will have an impact on overall demand for computing resources, The Paper reported today, citing industry insiders.
DeepSeek said it has experienced multiple service interruptions this year because of an imbalance between the supply and demand of computing power and insufficient system resilience. The most recent outage occurred on July 22 and lasted about four hours.
Editor: Emmi Laine
