"DeepSeek-Driven Realignment" Chinese AI Firms Slash Prices in Succession, Securing Competitiveness With Domestic AI Chips

DeepSeek V4 Pro usage surges after 75% price cut
Other Chinese AI firms recalibrate pricing strategies on the back of domestic chips
China’s rapidly expanding indigenous AI chip supply chain narrows Nvidia’s room to maneuver

China’s generative artificial intelligence (AI) frontrunner DeepSeek is reshaping the market with its pricing strategy. Major Chinese AI companies are following DeepSeek’s lead with successive price-cutting campaigns, rapidly unsettling an AI competitive landscape that had long been dominated by U.S. players such as OpenAI and Anthropic. Market observers point to China’s rapidly maturing domestic AI chip supply chain as the backdrop that has enabled Chinese firms to pursue such drastic strategic shifts.

DeepSeek’s Aggressive Price Cut

According to the South China Morning Post (SCMP) on the 8th, DeepSeek said last month that it would offer its latest flagship AI model, V4 Pro, at a price equivalent to 75% of its launch price. The move makes permanent a promotional discount introduced after the V4 model was released in April. As a result, the official application programming interface (API) price for V4 Pro has fallen to $0.0036 per 1 million input tokens and $0.87 per 1 million output tokens. OpenAI’s latest model, GPT-5.5, is priced at $5 per 1 million input tokens and $30 per 1 million output tokens.

The price-competitiveness gap between Chinese and U.S. AI models has widened sharply as a result. According to AI model performance evaluation and benchmarking firm Artificial Analysis, DeepSeek V4 Pro recently ranked first globally in “AI performance efficiency per dollar.” The cost of running V4 Pro through an AI index test, which compares multiple performance metrics across AI models, was about $268. GPT-5.5 and Anthropic’s Claude Opus 4.7, the latest models from OpenAI and Anthropic, cost 12 times and 19 times more, respectively, to perform the same task. With the global shortage of AI computing infrastructure driving up the cost of cutting-edge U.S. AI models, DeepSeek has emerged as an attractive alternative.

The dramatic price shift has, in practice, driven an immediate surge in market demand. Earlier this month, DeepSeek ranked first in weekly token usage on OpenRouter, the world’s largest AI platform. Chinese AI had overtaken OpenAI and Anthropic to claim the top spot. The shift is particularly striking given that Anthropic’s Claude Opus and Sonnet held the first and second spots as recently as early this year. OpenRouter is a kind of “AI shopping mall” that gathers numerous AI models from around the world in one place and allows users to freely choose among them. Developers use OpenRouter to connect AI models suited to each situation to agent programs.

Price Competition Heats Up in China’s AI Market

Other Chinese technology companies are also following DeepSeek with a cascade of price reductions. Xiaomi, the smartphone and electric vehicle manufacturing giant, moved most quickly. The company introduced an extreme measure cutting API usage fees for its flagship AI model, MiMo-V2.5, by as much as 99% from previous rates, immediately triggering an explosion in traffic. After the price-cut policy was announced, user adoption of Xiaomi’s MiMo-V2.5 and MiMo-V2.5-Pro models surged, with MiMo-V2.5 in particular climbing to around sixth place in OpenRouter’s popularity rankings.

A diversification of pricing systems is also taking shape. MiniMax, one of China’s leading AI unicorns, unveiled its next-generation flagship model, MiniMax M3, on the 8th and introduced a new hybrid pricing structure that combines the existing pay-as-you-go model based on token usage with subscription plans ranging from $7.24 to $69.28 per month. The strategy appears designed to improve cost predictability for high-frequency users while offering low-frequency users the flexibility of usage-based billing, thereby strengthening lock-in effects.

The industry sees the aggressive pricing policies of China’s AI sector as a product of the country’s domestic supply-chain ecosystem. With U.S. restrictions making it impossible to procure Nvidia graphics processing units (GPUs), the use of domestic AI chips has expanded sharply. One industry source said, “When DeepSeek released the V4 model last month, it had priced the service relatively high due to constraints in computing resources,” adding, “It appears to have finalized the price cut as the timing of mass Huawei chip supply became visible last month.” The source continued, “The same applies to other Chinese AI firms,” adding, “Chinese infrastructure that grew rapidly because of U.S. sanctions is now threatening U.S. Big Tech.”

China Accelerates AI Chip Self-Reliance

This trajectory toward ecosystem independence is expected to accelerate further. Alongside Huawei, numerous Chinese companies are stepping up their own AI chip development efforts. Alibaba, for instance, established semiconductor subsidiary T-Head in 2018 and began developing AI chips. In 2019, it unveiled Hanguang 800, its first AI inference chip, and applied it in practice to e-commerce search and recommendation systems as well as cloud services. More recently, Alibaba is said to be considering a separate corporate spin-off of T-Head and a potential initial public offering (IPO).

Baidu began its Kunlun project in 2011 and unveiled its first-generation chip in 2018. In 2021, it spun off its AI chip division into a separate entity, Kunlunxin, and began mass production of Kunlun II, based on a 7-nanometer process. Kunlun II is currently used in Baidu’s generative AI models, cloud data centers and autonomous driving business. Tencent is reported to have pursued its Zixiao AI inference chip project since 2021 and recently announced that it had built infrastructure offering full support for Chinese-made AI chips. ByteDance is also pushing ahead with in-house chip development after hiring large numbers of AI semiconductor design personnel, though it is still known to use Huawei Ascend chips alongside products from external suppliers.

Against this backdrop, Nvidia’s market share in China, despite its dominance of the global cutting-edge AI market, is falling by the day. Nvidia Chief Executive Jensen Huang said at Taiwan’s Computex event in May last year that the company’s China market share had fallen to 50% from 95% four years earlier, arguing that U.S. export restrictions on China had accelerated the growth of Chinese competitors. Reuters also cited data from market research firm IDC as saying that Nvidia’s share of China’s AI accelerator market by shipments was only about 55%, or 2.2 million units, last year. Chinese vendors accounted for 41%, or 1.65 million units, over the same period.