DeepSeek doubles peak-hour API price after sparking China AI price war
Synopsis
DeepSeek — the company that triggered China's AI price war with a 75% discount in May — is now doubling its own API rates during peak hours, exposing the unsustainable economics of racing to the bottom on model access pricing.
Key Takeaways
DeepSeek will double API prices during peak hours ( 9am–noon and 2pm–6pm Beijing time ) effective from the notice issued on Monday, 30 June 2026 .
Peak-hour cost for V4 Pro rises to 12 yuan (US$1.77) per million output tokens, up from the standard rate of 6 yuan .
The surcharge covers all V4 models and was communicated via subscriber email and a notice on DeepSeek 's website.
DeepSeek cited 'better distribution of resources and [to] enhance service stability' as the rationale for the increase.
The move reverses course from DeepSeek 's own May 2026 decision to cut V4 API prices by 75 per cent , which forced rivals ByteDance and Tencent Holdings to match the discounts.
DeepSeek, the Hangzhou-based Chinese artificial intelligence unicorn, is introducing peak-hour surcharges on its API services — a sharp reversal from the aggressive 75 per cent discount it announced in May 2026 that ignited a fierce domestic price war among AI providers.
What DeepSeek announced
The company notified service subscribers via email on Monday, 30 June 2026 that it will double the cost of accessing its V4 AI models through its application programming interface during peak hours — defined as 9am to noon and 2pm to 6pm Beijing time. A separate notice published to DeepSeek's website confirmed the surcharge applies to all V4 models.
Under the revised structure, peak-hour pricing for V4 Pro will rise to 12 yuan (US$1.77) per million output tokens, up from the standard off-peak rate of 6 yuan. DeepSeek attributed the change to the need for 'better distribution of resources and [to] enhance service stability,' according to the subscriber email.
Why it matters
DeepSeek was itself the catalyst for the pricing pressure it is now stepping back from. When the company slashed V4 API access fees by 75 per cent in May, rivals including ByteDance and Tencent Holdings were compelled to match the cuts, compressing margins across the sector. The peak-hour surcharge signals that even the company that started the race to the bottom is confronting the infrastructure costs of sustaining it.
AI companies typically monetise model access through token-based API pricing — charging developers per million tokens processed — making load management during high-demand windows a direct cost variable. By charging more during peak periods, DeepSeek is effectively shifting capacity costs onto its heaviest users.
The competitive backdrop
China's AI services market has seen relentless price compression in 2026, with major platforms treating model access as a loss-leader to lock in developer ecosystems. ByteDance and Tencent Holdings both followed DeepSeek's May discount, making the sector-wide margin squeeze a direct consequence of DeepSeek's own earlier move. The surcharge introduces a two-tier pricing dynamic that other providers may now feel pressure to replicate.
What's next
Whether ByteDance and Tencent Holdings follow with their own peak-hour pricing adjustments will be the immediate signal to watch. If rivals hold flat rates to maintain competitive differentiation, DeepSeek risks ceding developer loyalty at precisely the moment the market is consolidating around a handful of dominant model providers.
Point of View
It suggests GPU capacity costs are outpacing the revenue that flat-rate token fees generate. The two-tier structure DeepSeek is now piloting mirrors what cloud hyperscalers have long used to manage demand — a maturation that cuts against the narrative of Chinese AI providers as purely price-aggressive insurgents. If ByteDance and Tencent Holdings hold their discounted rates, the competitive pressure on DeepSeek's developer base could be significant.
NationPress
30 Jun 2026
Frequently Asked Questions
What is DeepSeek's new API pricing for peak hours?
DeepSeek will charge 12 yuan (US$1.77) per million output tokens for its V4 Pro model during peak hours — double the standard off-peak rate of 6 yuan . Peak hours are defined as 9am to noon and 2pm to 6pm Beijing time , and the surcharge applies to all V4 models.
Why is DeepSeek raising its API prices?
DeepSeek said the increase is needed for 'better distribution of resources and [to] enhance service stability,' according to the email sent to subscribers. The move follows infrastructure strain caused by high demand after the company cut prices by 75 per cent in May 2026 .
Did DeepSeek start the China AI price war?
Yes. DeepSeek triggered the price war in May 2026 when it announced a permanent 75 per cent discount on V4 API access. Rivals ByteDance and Tencent Holdings were forced to follow suit, compressing margins across China 's AI services sector.
How does API token pricing work for AI companies?
AI companies sell access to their models via APIs and typically charge fees based on token usage — usually per million tokens processed. Tokens represent chunks of text, and costs scale with the volume of data sent to and received from the model.
Will ByteDance and Tencent also raise their AI API prices?
It is not yet confirmed whether ByteDance or Tencent Holdings will introduce similar peak-hour surcharges. Their response will be a key indicator of whether the sector-wide discount cycle is ending or whether they will hold lower rates to gain a competitive edge over DeepSeek .
