Approaching.AI Raises 1.3 Billion Yuan in a Series B to Sell AI Tokens at Industrial Scale

A Chinese company that sells raw AI computing capacity the way a utility sells electricity has raised another large round. Approaching.AI, known in China as Qujing Technology, announced a Series B of more than 1.3 billion yuan, roughly 180 million dollars at current exchange rates. The round comes only a few months after a Series A completed in July, and brings total fundraising this year to 2.4 billion yuan, about 337 million dollars.

The round was co-led by a consortium of six investors including ICBC Financial Investment, Zhongguancun Science City, Xiaoshan Capital, Xinglian Capital and Qianjiang Century City, with participation from more than a dozen others including Yunchuang Capital, CITIC Juxin and Zhuhai Technology Industry Group. Existing shareholder Shangshi Capital doubled down. Several of the names are state-linked, which is worth noting: inference capacity in China is increasingly treated as infrastructure, and infrastructure attracts state money.

What the company sells is what its chief executive Ai Zhiyuan calls high-quality AI tokens: managed inference capacity aimed at production workloads, where the pitch is low latency, high throughput and stability rather than raw model capability. The company reports that daily production of these tokens has grown more than a hundredfold in six months and now exceeds one trillion per day, and that revenue this year has reached nearly ten times the whole of last year. Those are company figures and have not been independently audited.

The capital is earmarked for expanding token production capacity, upgrading its ATaaS platform, and continued work on inference infrastructure. The company says its technical route is deliberately narrow: few models, deep optimization, high quality. It lists in-house work on heterogeneous inference across chip types, a storage-for-compute trade it describes as exchanging memory capacity for arithmetic, and scheduling techniques it claims are global firsts. As with the revenue numbers, those claims come from the company itself.

The investor list doubles as the partner list. Approaching.AI says it works with SenseTime, Paratera, Hongxin Electronics, Wangsu Technology, MetaX and Tianshu Zhixin on building token factories on domestic Chinese chips, part of a broader push to make domestic accelerators usable for serious inference workloads. That is a different bet from the American neocloud model, which runs almost entirely on NVIDIA silicon.

The funding lands in a week when the American market is openly questioning whether AI capex will ever pay back, with Apollo economists counting zero productivity payoff outside the largest tech firms. The Chinese version of the argument is slightly different: capital is pouring into the layer that turns models into metered output, on the theory that whoever delivers reliable tokens at the lowest cost per unit captures the margin. Approaching.AI is one of the largest private bets on that thesis, and its next test is whether ten-times revenue growth survives contact with a market full of competitors doing the same thing.