AI Infrastructure Firm Secures 2.3 Billion Yuan in Series C Funding

The artificial intelligence infrastructure company Wuwen Xinqiong has completed a Series C funding round valued at 2.3 billion yuan. This financial injection brings the company's total cumulative funding since its inception to 4.3 billion yuan. The amounts for both the single round and the cumulative total set new records for the highest figures in the Chinese AI-native infrastructure sector. ITHome reported the story, citing information from the Securities Times.

The company was established in May 2023 as a technology transfer enterprise originating from Tsinghua University and Shanghai Jiao Tong University. It was incubated by the NICS-EFC laboratory team within the Department of Electronic Engineering at Tsinghua University. The firm distinguishes its business model by focusing on the middle layer of the AI ecosystem, specifically addressing engineering challenges such as heterogeneous computing power scheduling, software-hardware collaborative optimization, and large-scale cluster utilization. It does not develop its own GPU or AI accelerator hardware, nor does it directly develop underlying large models. Instead, it proposes a technical paradigm connecting multiple models with various chips.

Operational metrics indicate that the company's computing power distribution center has deployed resources reaching 37,000 petaflops, covering 16 types of mainstream chips. Its heterogeneous chip mixed training platform, which supports thousand-card scale and has ten-thousand-card extensibility, has achieved mixed training of large models using four domestic chips alongside two imported chip brands. The maximum computing power utilization rate reached 97.6 percent. In cross-domain scenarios, the company collaborated with China Telecom to complete the first domestic large-scale cross-domain mixed training technical verification over a distance exceeding 4,000 kilometers, facilitating the efficient training of a 30-billion parameter model.

The company's token factory, which corresponds to its Agentic MaaS large model service platform, utilizes a cross-cluster heterogeneous PD separation architecture. This architecture reduces the time to first token latency by 51.5 percent and lowers the cost per token by 37.5 percent. Over the past year, the token factory has driven a tenfold decrease in inference costs. The firm has established partnerships with leading large model enterprises and provides industry solutions covering entertainment, gaming, medical health, legal services, and energy sectors.

Regarding its public market ambitions, Bloomberg reported this month that the company has confidentially submitted an IPO application to the Stock Exchange of Hong Kong in recent months. The plan involves raising several hundred million dollars, with a target listing as early as the first half of 2027. Previous funding rounds include a Series A of nearly 500 million yuan in September 2024, a Series A+ of nearly 500 million yuan in November 2025, and a round exceeding 700 million yuan announced in May 2026. The investor base includes state-owned institutions, industrial investors, and financial investment firms.