Approaching.AI Banks $195M Series B as China's Token Race Heats Up

Approaching.AI has closed a Series B round of more than 1.3 billion yuan — roughly $195 million — bringing its total fundraising in 2026 to 2.4 billion yuan, or about $359.9 million, the Chinese AI token producer announced. That's four major rounds in a single year, roughly one every two months, each larger than the last.

The round was led jointly by Zhecai Innovation, Xiaoshan Capital, ICBC Investment, Zhongguancun Science City, Xinglian Capital and Qianjiang Century City. Follow-on investors included Yunchuang Capital, Hunan Gaoxin Zongheng, Jiantou Huake, Shenwan Innovation Investment, CITIC Juxin, Zhuhai Technology Industry Group, SenseTime, Wangsu Technology, Gansu Shuzhi, Changjinkong, Yuanzhi Future, Qingzheng Xingzhu and Xiaoka Capital. Existing shareholder Shangshi Capital increased its stake.

Who's actually writing the checks?

Look closely and the investor list reads like a map of state capital. Zhecai Innovation is a financial platform owned by Zhejiang province. ICBC Investment sits inside Industrial and Commercial Bank of China's financial asset investment system. Zhongguancun Science City represents Beijing's Haidian district, while Xiaoshan Capital and Qianjiang Century City are rooted in Hangzhou's Xiaoshan district. Capital from Hunan, Zhuhai and Gansu, plus investors tied to Shenwan, CITIC and China Construction Bank, fill out the follow-on roster. Mapped geographically, the base looks like a blueprint for AI token production capacity.

A two-year climb from seed money

The company was founded in December 2023. Zhenzhi Venture Capital provided seed funding in August 2024. In February 2025 it raised a tens-of-millions-of-yuan angel round led by Hillhouse Venture Capital and Zhipu's Z Fund. Then 2026 accelerated: an angel-plus backed by Paratera in February, a several-hundred-million-yuan pre-A in May, a Series A in July and now a Series B in October.

What does this change for AI infrastructure buyers?

Approaching.AI sells what it calls high-quality AI tokens — anchored to production-grade scenarios with high intelligence, low latency, high speed and stable delivery at controllable cost. The company argues the industry's early phase chased token volume over quality, and that as agent-driven deployment and large models take on core production processes, a quality tier is emerging.

It runs two commercial models: selling tokens from its own compute to model developers, internet platforms and enterprise clients, and a lighter business designing, building and co-operating token factories for customers with their own compute — the deployment focus in the second half. The numbers are striking. June revenue alone exceeded all of 2025, and this year's business revenue is nearly ten times last year's. Since the Spring Festival, token production efficiency per unit of compute has risen more than threefold, while average daily output of high-quality tokens has grown more than a hundredfold in six months to the trillion level.

On technology, the company sticks to fewer models and deeper optimization, engineering storage-for-compute substitution and virtual-real isomorphism. It says it has run trillion-parameter leading models on Chinese-made chips in real production, with cost-effectiveness it claims surpasses the world's most advanced compute.

Partnerships include SenseTime, Paratera, Hongxin Electronics, Wangsu, Moore Threads, MetaX and Iluvatar CoreX. SenseTime and Wangsu both invested in this round. SenseTime is building a token factory with daily output in the hundreds of trillions with the company; Wangsu is pairing its global edge network with inference optimization. Zhu Lifeng, chairman of Xiaoshan Capital Group, called the company "a benchmark project for Xiaoshan's AI infrastructure push," with a 10,000-card AI token factory under construction.

CEO Ai Zhiyuan said the real barrier in the compute race is energy efficiency per unit and commercial delivery capability, not scale, and that tenfold growth in six months validates the fewer-models approach. Proceeds will fund capacity expansion, the ATaaS platform upgrade, inference R&D and delivery systems.

The sector is moving toward public markets: Hong Kong filings show SiliconFlow applied to list on June 30, 2026, and Jiuzhang Cloud filed a prospectus on September 29, while Infinigence AI is reported to have filed confidentially. Policy is loosening too — the STAR Market's fifth listing standard expanded to AI large-model companies on June 17, and the first science and technology innovation corporate bond dedicated to token compute factories was issued on the Shanghai Stock Exchange on July 10.

The open question is profit. AI infrastructure companies generally face growing revenue without growing profit. Approaching.AI says it has crossed the cost line overall. Whether that holds as it replicates across more compute centers, chip combinations and customers is the test ahead.