China Processes 140 Trillion AI Tokens Daily as New 'Token Economy' Reshapes Its Tech Landscape

China now has an official word for token: ciyuan (词元). It was unveiled by Liu Liehong, head of China’s National Data Administration, at a State Council press conference in March — a moment that crystallized just how central AI inference has become to the country’s economy. If tokens are the new unit of computation, China is arguing it has become one of the world’s largest producers and consumers of that unit.

The numbers Liu cited are remarkable. China processed 100 billion tokens per day at the start of 2024. Today, that figure stands at 140 trillion — a 1,400-fold increase in roughly two years. During March 2026, China ranked among the top users of large AI models globally, with weekly token usage leading all other countries for three consecutive weeks.

From Hype to Revenue: The Agent Inflection Point

The numbers are not just a vanity metric for infrastructure. “Since late January this year, some model companies have achieved revenue in just 20 days that exceeds their total earnings for all of 2025,” Liu said in his statement. That acceleration reflects a shift that has been building since late 2025: the emergence of AI agents as the primary consumption driver, rather than chatbot conversations.

Where a single chatbot query might consume hundreds of tokens, an agent completing a multi-step task — researching a market, drafting a report, executing a booking, managing an email thread — can consume millions. The explosion of AI agent applications in China’s enterprise and consumer markets has transformed token economics in the same way that streaming video once transformed bandwidth economics.

China’s equivalent of the agent craze spans industries. In financial services, agent frameworks are handling loan application screening, fraud detection, and customer service escalations at major state-owned banks. In e-commerce, agents are managing end-to-end customer journeys for merchants on Taobao and JD.com. In healthcare, agents are serving as preliminary diagnostic assistants at tens of thousands of clinics.

The IPO Wave: MiniMax, Zhipu, and Biren

The commercial intensity of China’s AI boom has found its most visible expression in Hong Kong’s equity markets, which have experienced their best IPO conditions in five years, driven overwhelmingly by AI and tech listings.

Biren Technology set the tone by becoming the first mainland Chinese GPU developer to complete a Hong Kong IPO, raising $624 million in a January 2026 listing. Biren’s BR100 chips, positioned as domestic alternatives to Nvidia H-series GPUs for inference workloads, have found significant traction among Chinese cloud providers subject to U.S. export restrictions on advanced Nvidia products.

Zhipu AI listed a day before MiniMax in January, raising capital in one of the most anticipated Chinese AI IPOs in years. Zhipu’s business is concentrated among Chinese state-owned enterprises and major financial institutions, where its GLM model family has strong adoption. Zhipu reported 724 million yuan ($104.8 million) in 2025 revenue, up 132% year-on-year — but also disclosed total losses of 4.7 billion yuan ($680 million), driven by R&D expenditures that climbed 45% in the same period. Shares debuted with a modest 13% gain but have since surged more than 570% from the IPO price.

MiniMax was the more spectacular story. The company raised 4.8 billion Hong Kong dollars ($620 million) in its January 9 listing and doubled on debut, outshining its rival by every market metric. MiniMax reported $79 million in 2025 revenue — a 159% year-on-year increase — with 70% of revenue coming from international markets, particularly Singapore and the United States. Like Zhipu, MiniMax is burning cash aggressively, posting an adjusted net loss of $250 million. But for investors, the trajectory of the revenue curve has mattered more than the losses. Shares have risen more than 470% since the IPO, at one point briefly eclipsing Baidu’s market capitalization.

Both companies launched updated models just before their IPO hearings — a sequencing that speaks to the pace at which Chinese AI labs are iterating their foundation models even as they navigate capital markets.

Pony AI Goes Global

No company better illustrates the cross-border ambition of China’s AI moment than Pony AI. The autonomous vehicle company, which listed in the United States in 2024, launched what it billed as Europe’s first commercial robotaxi service in Zagreb, Croatia in early April 2026, partnering with Uber and local operator Verne.

Pony is targeting a fleet of more than 3,000 vehicles globally by end of year — up from roughly 1,000 today — with deployments across 20 cities, nearly half of which will be in overseas markets. The company has developed its seventh-generation robotaxi hardware (Gen-7), which it is deploying through a joint fleet with Chinese mobility operator Chenqi. Its PonyWorld 2.0 physical AI engine, announced this week, adds self-improving simulation capabilities — meaning the fleet can train on real-world data to improve performance without requiring manual intervention.

The Zagreb launch is a meaningful milestone not just for Pony but for Chinese tech broadly. European cities are notoriously conservative about autonomous vehicle deployments, and a commercial service — even in a small market — establishes proof of concept that could accelerate rollout across the continent.

The Geopolitics of Ciyuan

China’s decision to give tokens a Chinese name and build national accounting around them is a deliberate act of technological sovereignty. By framing AI inference in indigenous vocabulary and tracking it as a national economic indicator, Beijing is signaling that it views the token economy not as an American-invented abstraction but as critical digital infrastructure subject to Chinese governance.

This has practical implications. China’s model of AI development — domestically controlled compute, state-coordinated data access, and a clear alignment between commercial AI labs and national strategic priorities — differs fundamentally from the U.S. model. The two are now in overt competition not just in the frontier lab rankings but in the unglamorous but economically decisive question of which country’s AI infrastructure will power the global majority of token processing by 2030.

For now, the raw scale figures favor the United States at the frontier model level: OpenAI, Anthropic, and Google collectively run inference infrastructure that likely exceeds China’s in raw throughput. But China’s 140-trillion-token-per-day figure — driven by a vast, AI-hungry domestic market of 1.4 billion people and increasingly aggressive export strategies from companies like MiniMax and Pony — means the gap is narrowing faster than most Western analysts expected.

The token is the new barrel of oil. China has decided it intends to be an OPEC-level player.