China AI firm DeepSeek raises V4 prices sharply
2026-08-18
Usage fees raised by up to 12 times
Chinese artificial intelligence (AI) startup DeepSeek raised the usage fees for its flagship model V4 on the 17th. When it launched the preview version in April, it had said it planned to cut prices in the second half of 2026, but that outlook appears to have been derailed by a shortage of domestically sourced semiconductors.
The price per 1 million input and output tokens for V4 rose by as much as 12 times. In addition to raising the base fee, DeepSeek doubled the unit price during daytime hours in China, when usage is concentrated. The impact of the price increase is significant for a company that has marketed itself on high performance and low prices.
Heavy traffic turns into supply constraints
Signs of the price hike surfaced on the 4th. OpenCode, a tool that develops software autonomously using multiple AI models, warned that day of possible errors, saying V4 Flash was seeing unprecedented levels of access and lacked processing capacity.
V4 Flash was made available for testing on July 31 ahead of the main release. It offers stronger coding capabilities than the existing preview version, but also places a heavier load on servers. The surge in attention and usage temporarily overwhelmed processing capacity.
DeepSeek had also warned on the 6th that it would raise prices soon and that the increase would be substantial. AI services in China are generally inexpensive, and DeepSeek had been among the lowest-priced providers.
On the 13th, it announced the official price increase and at the same time launched the full versions of V4 Flash and the higher-performance Pro model. Both are positioned around coding performance. The company faces growing pressure to boost revenue and expand computing resources such as semiconductors and servers.
Expansion plans clouded
In April, when it released the V4 preview version, the company said it planned to sharply lower the price of V4 Pro in the second half of 2026. It had expected Huawei Technologies, the telecoms equipment giant, to expand supplies of servers equipped with its in-house AI chips.
That outlook, however, proved wrong. Chief Executive Liang Wenfeng reportedly told an investor briefing that the issue lies in Huawei's production capacity. He also said output is limited and that the allocation for DeepSeek is far smaller than that for major Chinese internet companies.
Moonshot AI, another Chinese AI startup, began offering Kimi K3 in mid-July, saying it approached the performance of U.S. models, but halted new registrations a few days later because of a shortage of computing resources.
China's Nanfang Metropolis Daily reported on the 7th, citing internet users, that Chinese AI is catching up with the United States in performance, but still has room for improvement in usability and processing capacity.
DeepSeek plans to build a large data centre in Inner Mongolia in China and is also working on developing its own semiconductors. Both are aimed at securing computing resources and both carry heavy investment burdens. Raising funds and increasing fee income are essential.
The U.S. government restricts exports of AI chips to China, and semiconductors in China are also in short supply. Many people use virtual private networks (VPNs) to access overseas AI services that are otherwise restricted domestically. If price hikes and glitches continue, users may drift away.
Enjoyed this article? Share it with your network!