China’s artificial intelligence industry is locked in a price war of historic proportions. The latest salvo comes from Xiaomi, whose newly launched “MiMo” language model is priced at just $0.10 per million input tokens, a figure so low it has prompted analysts to question whether the economics of the Chinese AI industry are sustainable. The development, reported by the Seoul Economic Daily, marks a new extreme in a competitive dynamic that has been reshaping the global AI market since early 2024.
For context, OpenAI’s latest flagship model is priced at $3 per million input tokens, with output tokens costing $15. The gap between the cheapest Chinese model and the leading US alternative has now widened to 30-170 times, depending on the model and token type being compared. DeepSeek and Alibaba’s Qwen, two of the most widely used Chinese models, are priced at approximately $0.14 per million input tokens, still a fraction of their American counterparts.
The Origins of the Price War
The current price war traces its origins to May 2024, when DeepSeek launched its V2 model at 1 yuan (approximately $0.14 ) per million tokens. The move was a deliberate provocation aimed at challenging the pricing power of established players such as Baidu, ByteDance, and Alibaba. The response was swift and dramatic: all three companies followed within weeks, slashing their API prices by 80-97 percent or offering their APIs entirely for free.
What began as a competitive skirmish has since escalated into an industry-wide race to the bottom. More than 200 large language models in China are now locked in active price competition, according to the Seoul Economic Daily analysis. The result is a market where the marginal cost of AI inference is approaching zero for end users, even as the underlying computational costs remain substantial.
The Chinese government has inadvertently accelerated this dynamic by subsidizing 50 percent of power costs for AI data centers that use domestic chips from Huawei or Cambricon. This subsidy effectively lowers the floor price at which Chinese AI companies can profitably operate, enabling them to undercut foreign competitors on price while maintaining viable margins.
The Global Implications
The price differential between Chinese and US models is having profound effects on the global AI market. As we reported recently, Kimi K2.6 from Moonshot AI topped the OpenRouter global leaderboard, while five of the top ten most-used models on OpenRouter by traffic in the week of May 4-8 were Chinese. Tencent’s Hy.3 model led the rankings with 3.74 trillion tokens processed per week, followed by Moonshot AI’s Kimi K2.6 at 1.78 trillion tokens per week.
This global adoption is being driven in large part by price. For developers in the Global South, startups operating on tight budgets, and enterprises running high-volume inference workloads, the difference between $0.10 and $3.00 per million tokens is the difference between a viable product and an unaffordable one. Chinese models are not just competitive on price; they are increasingly competitive on capability, making the value proposition compelling.
The implications for US AI companies are significant. OpenAI, Anthropic, and Google have historically been able to command premium prices based on their performance leadership. As the capability gap narrows. The Stanford 2026 AI Index found that the leading US model is now only 2.7 percentage points ahead of the best Chinese model on the Arena Leaderboard, the price premium becomes increasingly difficult to justify.
The Sustainability Question
The central question hanging over China’s AI price war is whether it is sustainable. The Seoul Economic Daily analysis highlights a growing concern among industry insiders: that the race to zero is destroying the economic foundations of the industry.
Li Qiang, Vice President of Tencent, offered a pointed analogy: “If tokens are compared to automobile fuel, ignoring the efficiency of the ‘engine’ and focusing only on fuel consumption will ultimately lead to user costs becoming so large that they will be shunned.” His broader argument is that token sales are a “non-sticky business” — customers have no loyalty to a provider and will switch immediately to whoever offers the lowest price. This makes it nearly impossible to build durable competitive advantages through pricing alone .
The subscription model, which many Chinese AI companies have been exploring as an alternative revenue stream, presents its own challenges. Heavy users — the power users who generate the most value for AI platforms — can consume hundreds of millions of tokens per month. At current pricing levels, these users can quickly collapse the revenue structures of subscription tiers that were designed for average consumption patterns.
(Related: ByteDance Tests Paid Subscriptions for Doubao as AI Monetization Race Heats Up)
The Strategic Pivot
Faced with the unsustainable economics of the token price war, some of China’s largest AI companies are beginning to pivot their strategies. Tencent, for example, is increasingly focusing on AI agent solutions and WeChat-integrated business-to-business cloud services, rather than competing purely on token pricing. This approach aims to move up the value chain, offering integrated solutions that are harder to commoditize than raw inference capacity alone.
This strategic evolution mirrors a broader trend in the global AI industry, where the initial phase of competition on raw model capability and price is giving way to a more complex landscape of application-layer differentiation, enterprise integration, and ecosystem lock-in. The companies that survive China’s token price war will likely be those that can successfully make this transition.
For now, however, the price war continues. With over 200 models competing for market share and government subsidies effectively lowering the floor, there is little near-term incentive for any individual company to unilaterally raise prices. The result is a market that is extraordinary in its scale and dynamism, but whose long-term economic viability remains an open question.