As the global race for artificial intelligence (AI) heats up, the AI ecosystem is being reframed around an unlikely and less-heard-about metric called tokens. These units of text, code, and data, which once were a back-end technical detail, are fast emerging as the basis of how artificial intelligence is priced, deployed, and scaled. As companies move from experimental chatbots to high-volume AI agents, token consumption is rising sharply, turning cost per token into a critical factor for developers and businesses alike.

According to recent industry data cited by the Financial Times (FT), Chinese AI companies are closing the gap in this area with significantly lower pricing models, and this change is starting to have an impact on the way in which the AI industry operates.

What is tokenisation in artificial intelligence and how do tokens work?

At the core of every generative AI system is a simple unit called a token, which is the smallest piece of data that a model reads, processes, and produces. These can be words, parts of words, punctuation marks, or even spaces.

For instance, if a user writes a prompt to be processed by the model, the model does not “read” the language in the way that humans do. It processes the language by predicting the next token in a sequence in order to produce a response to that prompt.

For instance, approximately 100 tokens equate to around 60 to 80 actual words in the English language.

Why are AI tokens being called the ‘currency’ of artificial intelligence?

Tokens are not just a technical concept, they are also how AI is priced, as most AI companies charge developers based on cost per token, meaning every prompt and response directly translates into a bill. This turns tokens into a unit of consumption, similar to how electricity is measured in units or cloud computing in usage time.

How China is gaining ground in global AI token pricing economics

Recent industry data cited by FT suggests Chinese AI firms are gaining share in global token usage, particularly among developers using cost-sensitive applications, and the advantage is largely cost-driven.

Chinese models are often priced significantly lower per token than their US counterparts, making them attractive for high-volume workloads, FT reported. This is especially relevant for AI agents and automation tools, where token consumption can scale sharply.

And majorly two structural factors are driving this:

• 
  Lower energy costs, supported by large investments in renewable power
• 
  More efficient model architectures, designed to reduce computational load

Additionally, US export restrictions on advanced chips have pushed Chinese companies to optimise software efficiency, indirectly improving cost per token. 

Can China sustain its advantage in AI token pricing over time?

China’s lead in token economics is currently there, but it is not without its constraints. While some models have faced capacity issues when usage spikes, highlighting infrastructure limits, there are also concerns among global users about relying on models hosted in Chinese data centres, FT said.

At the same time, US companies continue to dominate in high-performance models and enterprise-grade reliability.