Infinigence AI, the Shanghai-based AI infrastructure startup better known in China as Wuwen Xinqiong, has confidentially filed for an initial public offering in Hong Kong, Bloomberg reported Friday, citing people familiar with the matter. The company is targeting a raise of several hundred million US dollars and could list as early as the first half of 2027, though the plan is still being finalized and both the timing and the size could change.
The filing makes Infinigence one of the first Chinese "AI cloud" operators to head toward public markets. The company was founded in May 2023 out of Tsinghua University's NICS-EFC lab by Wang Yu, chair of the university's electronic engineering department, together with his former doctoral students Xia Lixue (CEO), Dai Guohao (chief scientist) and Li Boxun (CTO). Wang previously co-founded the chip startup Deephi, which Xilinx acquired in 2018.
According to Bloomberg and Chinese media recaps of the filing, Infinigence has raised about RMB 4.3 billion (roughly $600 million) since founding, at a pre-IPO valuation of RMB 14.3 billion (about $2 billion). Its backers read like a map of China's AI stack: Tencent, Baidu and fellow model developer Zhipu AI are strategic shareholders, alongside Alibaba, Xiaomi, Lenovo Capital, HongShan, Northern Light Venture Capital and GSR Ventures. Its most recent disclosed round, over RMB 700 million in May 2026, was co-led by Hangzhou state-backed investors and was described at the time as the largest financing for a Chinese AI-native infrastructure company that quarter.
What makes the story unusual is what Infinigence does not own: data centers. Unlike Western neoclouds such as CoreWeave or Nebius, which borrow billions against GPU fleets they hold themselves, Infinigence operates an asset-light orchestration layer. The company says it pools compute spread across dozens of third-party facilities and adapts workloads to more than 16 types of AI accelerators, including Huawei's Ascend chips and silicon from domestic challengers such as Moore Threads. Company figures cited in Chinese coverage put its reachable compute at more than 37,000 petaflops and claim a thousand-card heterogeneous training cluster reaches 97.6 percent utilization — numbers reported by the company and not independently verified.
The pitch lands on a real bottleneck in Chinese AI. Developers there can draw on a fragmented menu of domestic and imported chips, each with its own software stack, and model teams routinely re-port work for each vendor. Infinigence's founders have described the company as the refinery of the AI supply chain: it takes electricity and silicon as inputs and produces tokens as the output. China's daily token consumption crossed 140 trillion in March 2026, up more than 40 percent from the end of 2025, according to industry figures cited in Chinese media.
The listing plan follows a wave of Hong Kong debuts by Chinese AI companies this year — Zhipu AI and MiniMax both went public in January, and Moonshot AI has reportedly been preparing its own IPO — while in the US, CoreWeave's post-IPO trading has become the market's barometer for GPU-rental economics. A filing from an orchestration-first operator would give public investors their first close look at whether the software layer, rather than hardware ownership, is where the margins in AI compute settle.
That is the bet, and it is unproven. Infinigence's light balance sheet is precisely what its critics will probe: orchestrating other people's data centers means thinner control over capacity and cost, and the company's utilization figures come from its own marketing. But with China's token demand compounding and more than a dozen accelerator vendors to normalize, someone has to write the translation layer. Infinigence has convinced Tencent, Baidu and its other backers that it is that layer — and now it wants the public market to pay for scaling it.
Comments (0)
Log in to join the discussion
Log InNo comments yet