Recently, there's a quite interesting phenomenon: everyone suddenly started seriously discussing how to translate "token" into Chinese.

Tsinghua University professors wrote articles saying "Deciding on the Chinese translation of token is imminent," and related questions on Zhihu have garnered 250k views. But if you ask me, this is not really a translation issue.

Going back two or three years, the domestic AI community simply called it "token," and no one cared. Why is there suddenly a need for a Chinese name now? The direct reason is that after this year's Spring Festival, ordinary people first realized that tokens cost money. OpenAI's tools turned AI from chatting into real work, with each task burning through hundreds of thousands of tokens, and bills skyrocketing. Cloud providers also raised prices, with all billing units being tokens.

More importantly, tokens started appearing in places where they previously shouldn't have. Huang Renxun said at GTC that Silicon Valley is already asking during interviews, "How many tokens will this job consume," and even suggested including tokens in engineers' compensation. Sam Altman is more radical, believing that tokens will replace universal basic income. According to statistics from the National Bureau of Statistics, China's daily token consumption went from 100 billion at the start of 2024 to over 40 trillion by September 2025.

Once technical terminology enters cloud service bills, recruitment salary packages, and official statistics, it can no longer be called English. But this is not a translation problem; it’s a discourse power issue.

Academics have already named it "word element" (詞元) back in 2021, but no one paid attention because tokens weren’t valuable then. Now it’s different. The word "token" itself is a universal container; crypto people call it "tokens," security folks call it "tokens" or "tokens," AI folks call it "word elements." The same English word, depending on which direction the Chinese translation leans, belongs to whose territory.

Thus, a naming battle for "token" has begun. The most popular name is "Zhiyuan" (智元). The most active promoter is the new AI media outlet, Xin Zhiyuan (新智元). Think about it: if the Chinese name for token is set as Zhiyuan, their brand name would overlap with industry terminology, effectively turning every article discussing tokens into free advertising for them. Their own articles openly state: "We suggest translating token as an industry consensus: Zhiyuan, leaving the character 'new' (新) to us."

Bai Chuan Intelligence founder Wang Xiaochuan said that calling it Zhiyuan is quite good. He works on large models, so calling tokens Zhiyuan makes sense. Instead of billing units, each computation produces a "basic unit of wisdom." Selling tokens is selling traffic; selling Zhiyuan is selling intelligence. The valuation story is completely different.

Tsinghua professor Yang Bin proposed "Moyu" (模元), where "Mo" (模) corresponds to models, and whoever owns large models also controls the production rights of "Moyu." Names leaning toward models shift pricing power to model companies. Others advocate calling it "Fuyuan" (符元), returning to the fundamental definition in computer science: tokens are symbol processing units, unrelated to intelligence. Technically the cleanest, but the proposer is an independent technical author without a company platform or capital backing, so it has almost no voice in this discussion.

The direction of the name influences the industry narrative, and money flows accordingly. When Facebook rebranded as Meta, "metaverse" shifted from a sci-fi concept to a valuation story for a company. China consumes 180 trillion tokens daily, the world’s largest, but what this term is called, how it’s defined, and who defines it, are still unresolved. The world's largest token-consuming country hasn't even decided what to call what it consumes.

Huang Renxun did not participate in the Chinese naming discussion at GTC; he did something simpler: he raised a championship belt with "Token King" written on it, announcing that data centers are token factories. Whoever produces tokens defines them. He doesn’t care what the name is.

What truly matters is not which translation is better. Once the term "calorie" was established, the entire food industry’s pricing, labeling, and regulatory systems were built around it. After the definition of "traffic" was established in China’s telecom industry, operators billed based on traffic, competed on traffic, and designed packages around traffic, transforming the business model for over a decade. Tokens are now following the same path.

It has already become the billing unit for cloud services, the revenue metric for large model companies, and a core indicator for measuring AI industry scale at the national level. The VC circle is even discussing whether investment funds can be paid directly in tokens. Once a word becomes a measure of money, naming it is no longer translation; it’s minting currency.

Calling it "Zhiyuan" (智元) means minting rights belong to AI narratives—whoever tells the story of intelligence benefits. Calling it "Moyu" (模元) means minting rights belong to model companies—who has large models can print money. Calling it "Fuyuan" (符元) means minting rights return to technology itself, but technology cannot speak for itself. The "word element" (詞元) defined by academia in 2021 was ignored not because of poor translation, but because at that time, this "currency" was not valuable. Now it’s valuable, and everyone wants to carve their name on it.