An engineer (center) installs OpenClaw, an open-source AI agent, for users in Hangzhou, Zhejiang province, on March 22. [Photo for China Daily]

New type of 'currency' heats up smart tech market, reshapes firms' strategies

As tokens emerge as a new strategic resource in the artificial intelligence era, companies are urged to treat token consumption as a core cost metric and integrate it into budgeting and operational decision-making, industry experts said.

Their remarks came as Nvidia CEO Jensen Huang described tokens as "the new commodity" during a keynote address at the company's recent GTC conference, noting that they are becoming a form of currency underpinning recruitment, budgeting and productivity.

A token, in the field of AI, is the smallest unit of data processed by AI models. Generating a single Chinese character costs about 0.7 tokens, while a high-resolution image may use thousands and a 15-second video can consume around 300,000.

As tasks grow more complex, the growing importance of tokens is reflected in surging usage. The National Data Administration said that as of March, average daily token consumption in China had exceeded 140 trillion, up more than 1,000-fold compared with early 2024.

Further highlighting the trend, data from OpenRouter, a major AI model API aggregation platform, showed that the top four models globally by token usage during the week of March 16-22 were all developed by Chinese companies, including Xiaomi's MiMo-V2 Pro, StepFun's Step 3.5 Flash, MiniMax-M2.5 and DeepSeek-V3.2. Total weekly token usage by Chinese large models also increased by 56.91 percent week-on-week, surpassing their United States counterparts for the third consecutive week.

According to global market consultancy IDC, the rapid growth has been supported by a significant cost advantage. Benefiting from lower green energy costs, the unit price of tokens for mainstream Chinese models is only one-sixth to one-tenth that of overseas counterparts, IDC data showed.

"This means that if Chinese companies can fully leverage the domestic pricing advantage, every unit of their token budget will deliver greater purchasing power," said Lu Yanxia, research director at IDC China, adding that this is not only a way to improve efficiency, but also a key window for Chinese firms to gain an edge in global AI application competition.

"Just as companies in the industrial era had to budget for electricity, those in the AI era must learn to budget for tokens," Lu said.

She added that enterprises can strengthen refined token management by incorporating token usage into core financial metrics and linking budgets to business growth targets to ensure returns on investment.

In this regard, concepts such as "token budgets", "department-level token quotas" and "return on token" may emerge in corporate budgeting frameworks in the future, according to domestic media reports.

As advances in AI agents further amplify the importance of "token economics", the time has come to build an AI value assessment framework centered around tokens, said Zhang Ran, an analyst at the information technology management department of China CITIC Bank.

"The key lies in building a multidimensional framework, encompassing cost, efficiency, quality, returns and overall strategy," said Zhang.

In fact, major industry players are already moving in this direction. Alibaba Group this month established the Alibaba Token Hub (ATH) business unit, led directly by CEO Eddie Wu, setting tokens as a core strategic metric and requiring all operations to revolve around their creation, distribution and application.

Echoing this trend, Liu Jun, chief AI strategy officer at Inspur, said companies need to rethink system architecture with token cost reduction as a central goal.

"Previously, AI computing systems were designed to be large and comprehensive. But once we focus on lowering token costs, we need to identify system bottlenecks and rebuild architectures with a more streamlined design," Liu said.

For AI to truly scale and become widely accessible, token costs must fall by another order of magnitude from current levels, Liu added."Token costs are a core driver of competitiveness, directly determining the profitability of AI agents."