IT之家 · 智能时代 · 10/12/2026, 12:08:12
Infinigence AI Raises Record $310M Series C for AI Infrastructure: Heterogeneous Compute Scheduling and PD Separation Cut Costs
AI infrastructure firm Infinigence AI has completed a 2.3 billion RMB Series C round, bringing its total funding to 4.3 billion RMB and setting a new record for China's AI-native infrastructure sector. Specializing in the 'middle layer' via an M×N paradigm, the company enables heterogeneous chip mixed training and cross-domain scheduling; its proprietary PDD architecture reduces Time-To-First-Token latency by 51.5% and per-token costs by 37.5%. The company has also confidentially filed for an HKEX IPO, targeting a listing in the first half of 2027.
SOURCE COVERAGEFull Original Report
On October 12, according to Securities Times, AI infrastructure (AI Infra) company Infinigence AI has completed a Series C funding round of 2.3 billion RMB. Its cumulative funding since inception has reached 4.3 billion RMB, setting new records for both single-round and total financing in China's AI-native infrastructure sector.
Public information indicates that Infinigence AI was established in 2023 as a technology commercialization venture originating from Tsinghua University and Shanghai Jiao Tong University.
IT Home noted that Bloomberg reported this month, citing sources familiar with the matter, that Infinigence AI has confidentially filed an IPO application with the Hong Kong Stock Exchange in recent months. The company plans to raise hundreds of millions of dollars, targeting a listing as early as the first half of 2027.
Further Reading
Infinigence AI, officially known as Shanghai Infinigence Intelligent Technology Co., Ltd., was founded in May 2023. It was incubated by the NICS-EFC Laboratory team from the Department of Electronic Engineering at Tsinghua University and serves as a technology commercialization entity for both Tsinghua University and Shanghai Jiao Tong University. Regarding the team, co-founder and CEO Xia Lixue holds a PhD from Tsinghua University, having studied under Professor Wang Yu, Chair of the Department of Electronic Engineering at Tsinghua. He was selected for the AI2000 list of the world’s most influential scholars in artificial intelligence (Top 100 in the chip direction). Founder Wang Yu has long been engaged in research on intelligent chips and high-efficiency circuit systems; in 2016, he invested in Deephi Tech through knowledge transfer, which was acquired by Xilinx in 2018 and subsequently integrated into AMD. Another co-founder, Dai Guohao, serves as Chief Scientist and is a tenured-track associate professor at Shanghai Jiao Tong University. Unlike chip design companies such as Biren Technology and Enflame Technology, or large model enterprises like Zhipu AI and MiniMax, Infinigence AI follows a "middle layer" path: it does not develop its own GPU/AI accelerator hardware nor directly build underlying large models. Instead, it focuses on engineering challenges such as heterogeneous computing power scheduling, software-hardware co-optimization, and large-scale cluster utilization. It proposes an M×N technical paradigm connecting "M types of models" with "N types of chips." Specifically, its "One Center" computing power distribution hub has deployed access to 37,000 Petaflops of computing power, covering 16 mainstream chips. Its thousand-card scale heterogeneous chip mixed-training platform possesses ten-thousand-card scalability and has achieved large model mixed training between four domestic chips (Huawei Ascend, Tianshu Zhixin, MetaX, Moore Threads) and two imported chips (AMD, NVIDIA), reaching a maximum computing power utilization rate of 97.6%. In cross-domain scenarios, Infinigence AI collaborated with China Telecom to complete the first domestic verification of large-scale cross-domain mixed training over 4,000 kilometers. This involved implementing cross-domain training tasks for inference clusters from Hami, Xinjiang, to Shenzhen, Guangdong, achieving efficient and stable training for a 30-billion-parameter model. The "Back Factory" Token Factory corresponds to the Agentic MaaS large model service platform. The company pioneered a cross-cluster heterogeneous PD separation architecture (PDD), which reduces Time To First Token (TTFT) latency by 51.5% while lowering the cost per token by 37.5%. Over the past year, its Token Factory drove a 10-fold decrease in inference costs. In terms of customers and ecosystem, Infinigence AI has partnered with leading large model companies such as Zhipu AI and Moonshot AI. Its industry solutions cover sectors including entertainment and gaming, healthcare, legal services, and energy/power. Clients also include government and enterprise entities and research institutions such as China Mobile, China Unicom, Southern University of Science and Technology, and Migu Culture Technology. In previous funding rounds, Infinigence AI completed a Series A round of nearly 500 million RMB in September 2024, accumulating close to 1 billion RMB in funding just over a year after its establishment. In November 2025, it completed a Series A+ round of nearly 500 million RMB. In May 2026, it announced securing over 700 million RMB in funding, jointly led by Hangzhou High-Tech Financial Investment Group and Huiyuan Capital. The investor lineup spans state-owned capital institutions, industrial investors, and financial investment firms. State-owned capital institutions include Hangzhou High-Tech Financial Investment, Social Security Fund Zhongguancun Fund, Zhuhai Technology Industry Group, Xiamen Guoxing Investment, and Xuhui Capital. Industrial investors include Tencent, Baidu, Lenovo, and Xiaomi. Financial investment firms include Sequoia China, Qiming Venture Partners, Hongtai Fund, Dacheng Finance & Intelligence, and Shunwei Capital.