China gains edge in AI race through token pricing power

A recent analysis by the Financial Times highlights a shifting dynamic in the global artificial intelligence race, arguing that China is gaining a significant edge in the production and consumption of “tokens”—the core units processed by large language models.

Since February, Chinese AI developers such as DeepSeek and MiniMax have surpassed US competitors in token usage, according to OpenRouter data. Tokens—units of text, code or data—have become a key economic metric in AI, as companies charge per token, making them both a measure of adoption and a pricing battleground.

The shift is particularly important as AI agents—more advanced systems than traditional chatbots—consume vastly more tokens. While a chatbot might use around 30,000 tokens to summarise a text like Hamlet, an AI agent can require up to 20 million tokens for relatively small coding tasks. This surge in consumption is amplifying the importance of cost efficiency.

Chinese firms currently hold a strong pricing advantage. Companies such as MiniMax and Moonshot charge roughly $2–$3 per million output tokens, compared with about $15 for models like Claude Sonnet 4.5 developed by Anthropic. As Will Liang noted, even small price differences become critical when AI agents “burn through millions of tokens a day.”

This cost gap is already reshaping user behavior. Developers are increasingly adopting hybrid strategies—using cheaper Chinese models for routine workloads while reserving more expensive US systems for complex tasks. One developer cited in the report reduced projected daily costs from $900 to about $50 by switching primarily to Moonshot’s Kimi model.

The pricing advantage stems partly from China’s lower energy costs and heavy investment in renewables, alongside efficient model architectures such as “mixture-of-experts,” which reduce computing demands. Government policy is reinforcing this trend, with Beijing prioritizing “computing-electricity synergy” as a strategic objective.

However, constraints remain. Rapid growth has strained infrastructure, as seen when Zhipu AI faced service disruptions after a surge in demand for its GLM-5 model. Analysts stress that performance and reliability are as important as cost in determining long-term competitiveness.

US companies—including OpenAI and Google—continue to expand rapidly, but face mounting competition in price-sensitive segments. Meanwhile, Chinese tech groups such as Alibaba are moving to institutionalize their advantage, launching initiatives like a dedicated “Token Hub” to capitalize on what they see as the next phase of AI.

Despite this momentum, geopolitical risks could limit China’s gains. Concerns over data security and jurisdiction remain significant barriers, particularly for governments and regulated industries wary of relying on Chinese infrastructure.

By Tamilla Hasanova