By Hamza Faisal
Beijing-based AI infrastructure startup Approaching.AI has closed a Series B funding round of more than RMB 1.3 billion, roughly $195 million, betting that the next big money in artificial intelligence will be made not by training models, but by running them cheaply at scale.
The round, announced on October 26, was co-led by Zhejiang Finance Capital, Xiaoshan Capital, ICBC Investment, Zhongguancun Science City, Xinglian Capital and Qianjiang Century City. Participants included SenseTime and Wangsu Technology, while existing backer Shangshi Capital increased its stake. The company behind the Approaching.AI brand is Beijing Qujing Technology.
A fundraising sprint in a single year
The Series B caps an unusually fast fundraising run. Approaching.AI closed a pre-Series A round in May 2026 and a Series A in July, bringing its total fundraising for the year to RMB 2.4 billion. The company says several of its core operational indicators have grown more than tenfold in the past six months, though it has not disclosed revenue or profitability figures.
The new capital will go toward expanding what the company calls high-quality AI token production capacity, upgrading its ATaaS platform, short for Approaching.AI Token as a Service, developing underlying inference technology and new products, and strengthening commercial partnerships and delivery systems.
Why inference is the new battleground
Inference is the computing work of running a trained model: generating answers, summaries and code for real users. As AI applications spread through finance, healthcare and e-commerce, the cost of serving millions of requests has become a bottleneck, and investors are pouring money into companies that promise to shrink it.
Approaching.AI builds inference engines, custom accelerator chips and a compiler designed to cut the number of calculations needed per token, claiming up to a 40 percent reduction in energy consumption compared with conventional graphics processors. Its software-hardware platform also uses model quantization and dynamic batching to serve models efficiently, including on a customer's own premises. Its customers include large-model developers Zhipu and Moonshot AI.
A sector awash with capital
The round lands in the middle of a funding wave for Chinese inference infrastructure. SiliconFlow raised more than $275 million in June, and Infinigence closed a $316.3 million round in October, as investors bet that spreading AI applications will drive demand for token-serving capacity.
The open question is whether that demand converts into durable, profitable businesses. Approaching.AI has not published per-inference costs or margins, and the funding alone does not prove the model works. But with another large round closed in under a year, the company has bought itself considerable room to try.
By Hamza Faisal



