The economics of artificial intelligence in China are undergoing a violent restructuring. Following DeepSeek’s decision to permanently cut prices on its flagship V4-Pro model by 75 percent, a cascading wave of retaliatory price reductions has swept through the domestic tech sector, forcing rivals to slash application programming interface (API) costs and rattling the cloud infrastructure providers that host them.
The pricing shock, detailed in a new report by the South China Morning Post, illustrates how the “involution” dynamic we analyzed earlier this week is moving beyond the model developers themselves and beginning to squeeze the margins of the broader technology supply chain.
Xiaomi and Tencent Respond with 99 Percent Cuts
The immediate response from competitors has been aggressive. Smartphone and electric vehicle manufacturer Xiaomi was among the first to respond, slashing API costs for its MiMo-V2.5 model by up to 99 percent. The move produced an immediate surge in volume: usage of the MiMo-V2.5 and MiMo-V2.5-Pro models spiked, with the former climbing to sixth place on the US-based model marketplace OpenRouter. According to the SCMP, the model processed 1.7 trillion tokens in the seven days leading up to June 8, representing growth of more than 999 percent from the previous week.
The pressure is not limited to proprietary model developers. Third-party infrastructure providers are being forced to slash access fees to prevent developers from migrating to lower-cost alternatives. Tencent Cloud cut API prices for DeepSeek’s V4 models by as much as 97.5 percent this week, matching the discounts DeepSeek introduced.
This dynamic is unique to the open-weight model ecosystem. Unlike proprietary systems such as OpenAI’s GPT-4, open-weight models like DeepSeek V4 can be hosted by multiple vendors simultaneously. This creates fierce, commoditized competition among cloud providers and infrastructure companies, who must compete solely on the cost and speed of inference rather than the model’s underlying intelligence.
The Search for Sustainable Monetization
As headline token prices collapse toward zero, Chinese AI companies are experimenting with alternative monetization strategies to balance user growth with sustainable revenue.
AI unicorn MiniMax launched its next-generation flagship model, MiniMax M3, on June 1, pairing traditional token-based billing with subscription plans ranging from $7.24 to $69.28 per month. The transition highlights the friction inherent in changing business models: some users complained that token consumption under the new system was significantly higher than expected, exhausting monthly quotas prematurely. MiniMax was forced to apologize and grandfather existing users into their previous unlimited weekly access benefits.
“China’s AI price war is becoming more sophisticated,” noted Poe Zhao, a China tech analyst quoted by the SCMP. “Companies are increasingly experimenting with different charging models rather than relying solely on across-the-board price cuts. For enterprise users running complex workloads, the effective cost of completing a task may matter more than headline token prices.”
The Cloud Provider Squeeze
The most significant long-term consequence of the price war may be its impact on China’s cloud computing sector. Shares in Chinese cloud computing and server-rental companies have weakened since DeepSeek announced its permanent price cuts, reflecting investor concerns that collapsing inference costs will eventually erode demand for outsourced computing power.
The economic advantage of open models is stark. A November 2025 working paper by researchers from MIT, the Linux Foundation, and the Georgia Institute of Technology found that open models cost, on average, just 15.66 percent as much to operate as closed-source alternatives. As Chinese developers increasingly default to these cheaper open-weight architectures, the total addressable market for cloud inference revenue may shrink even as token volume explodes.
There is also a strategic dimension to the price cuts. He Baohong, chief engineer at the state-backed China Academy of Information and Communications Technology, noted at a digital economy forum last week that data availability remains one of the biggest constraints on advancing AI model performance. By dropping prices to near-zero, companies like DeepSeek and Xiaomi are effectively subsidizing user interaction to secure the massive volumes of real-world data needed to train their next-generation models.
The result is a market where survival requires deep pockets and a willingness to endure sustained losses. As the price war moves from the model layer to the cloud infrastructure layer, the question is no longer just which AI developer will win, but whether the infrastructure providers hosting them can afford the victory.