China's AI unicorn DeepSeek will double peak-hour fees for its V4 model API. DeepSeek, which sparked AI price competition in China with large discounts, has changed its pricing strategy by introducing surcharges for congested hours.

The South China Morning Post reported on June 30 that DeepSeek told subscribers in an email that it will raise V4 API prices during peak hours of 9 a.m. to noon and 2 p.m. to 6 p.m. Beijing time.

The increase applies across the V4 line. DeepSeek said in a separate notice that the new prices apply to all V4 models. As a result, the peak-hour fee for V4 Pro will rise to 12 yuan (about 2,750 won) per 1 million output tokens. The existing off-peak fee is 6 yuan (about 1,375 won). AI companies typically sell access to models via APIs and charge based on token usage.

The increase will apply from when the full version of V4 goes online in mid-July. V4 was first released in late April as a preview version, and the pricing system will change along with the shift to the full version.

DeepSeek cited service operations issues as the reason for the increase. The company said in the subscriber email that it needs "better resource allocation" and "stronger service stability". It framed the move as an effort to manage usage during high-demand hours rather than a simple price increase.

Some in the market view the decision as a signal that differs from the low-price competition in China's AI industry over the past several months. DeepSeek said in May it would permanently cut V4 API access prices by 75 percent, fuelling a price war. ByteDance and Tencent then moved to similar cuts, intensifying competition to secure market share.

Chinese AI companies' price-cut competition intensified further in June. In mid-June, ByteDance introduced its video generation model "Seedance 2.0 Mini" at 23 yuan (about 5,270 won) per 1 million tokens. That is about half the cost of its existing standard version. Xiaomi and Tencent had earlier cut prices for some AI models by as much as 99 percent and 70 percent, respectively.

Against that backdrop, DeepSeek's introduction of surcharges is being read as a move away from relying solely on across-the-board cuts to attract developers and corporate clients. It sets a clear price gap between peak and off-peak hours, aiming for both demand dispersion and operational stability for infrastructure.

A key question is whether Chinese AI companies will follow with time-of-day billing. If competition has until recently centred on price cuts to gain market share, more granular pricing policies that reflect service stability and resource allocation could spread. For now, DeepSeek will start testing a new fee system that reflects operating efficiency from the mid-July launch of the full V4 version, shifting from its position of leading price competition.