Skip to main content

基流科技

Category: AI Infrastructure

China's largest independent AI GPU cluster infrastructure provider, building full-stack AI computing clusters for large model training. 基流科技 was founded in 2023. The company is led by 胡效赫 (Hu Xiaohe). Based in Beijing (originally), Shanghai (HQ as of 2026), China. Team size: 101-500. Total funding raised: $308.3M. Latest round: Series D. Key investors include 光速光合 (Lightspeed China), 奇绩创坛 (MiraclePlus), 卓源亚洲 (Zhuoyuan Asia), 星连资本 (StarLink Capital / Zhipu AI ecosystem fund), 招商局创投 (China Merchants Group VC), 华泰数智 (Huatai Digital Intelligence), 国方创新 (Guofang Innovation), 上海国投孚腾资本 (Shanghai State Investment Futeng Capital), 河南投资集团汇融基金 (Henan Investment Group Huirong Fund), 启迪之星 (TusStar), 水木清华校友基金 (Tsinghua Alumni Fund), 英诺天使 (InnoAngel).

Founded
2023
Headquarters
Beijing (originally), Shanghai (HQ as of 2026), China
Team size
101-500
Total funding
$308.3M

Value proposition

Full-stack independent AI computing cluster builder that delivers 20%+ performance uplift, long-distance distributed training (30-1500km) with minimal compute loss, and industry-leading cluster reliability (99.97% SLA, <0.3% annual hardware failure rate).

Products and solutions

Galaxy HPAC (全栈开源自主高性能算力集群系统), Venus (AI算力操作系统/智算集群运行管理平台), Mercury (AI算力网络系统), GPU-RDMA集群管控通信框架, 高速无损网络交换机 (RoCE), 算力管控调度平台, 算力调优运维平台, 高速互联硬件, 端到端国产化AI算力集群产品

Unique value

China's largest independent AI GPU cluster provider — the only non-cloud-giant player with proven multi-10K GPU cluster deployment experience, delivering 20%+ performance improvement and cross-region distributed training capabilities.

Target customer

Large model labs (e.g.智谱AI/Zhipu AI), AI companies (e.g.商汤科技/SenseTime), telecom operators (China Unicom), data centers, local government-owned enterprises, and enterprises needing large-scale AI computing infrastructure.

Industries served

AI/ML model training, AI infrastructure, Cloud computing, Data centers, Telecommunications

How they differentiate

1) Independent (not owned by cloud giants) — largest independent AI cluster provider in China; 2) Proven 10K+ GPU cluster deployment experience (rare in China); 3) 20%+ performance improvement over standard clusters; 4) Long-distance distributed training (30km with 100% efficiency, 50km with 98-99%, 1500km demonstrated with China Unicom); 5) Full-stack domestic/国产化 capability; 6) Industry-leading reliability (99.97% SLA, <0.3% hardware failure rate).

Main competitors

Alibaba Cloud (阿里云), Huawei Cloud (华为云), Baidu AI Cloud (百度智能云) — the cloud giants' internal AI cluster divisions, also competes with traditional HPC/server vendors like 中科曙光 (Sugon) and 浪潮信息 (Inspur) in the broader AI infrastructure space.

Key partnerships

China Unicom (中国联通) — 1500km cross-region distributed training, 智谱AI (Zhipu AI) — strategic customer & investor via 星连资本, 商汤科技 (SenseTime) — customer, Multiple chip/switch vendors for domestic hardware adaptation, 上海国投孚腾资本, 国方创新, 河南投资集团汇融基金

Notable customers

智谱AI (Zhipu AI), 商汤科技 (SenseTime), 银盾云, 世纪互联 (21Vianet), China Unicom (中国联通), Multiple data centers and local government-owned enterprises

Major milestones

2023-02: Company founded, 2023: Released Venus AI computing OS, delivered first 1000+ GPU cluster, 2024: Delivered first 4000+ GPU cluster (45% MFU), first 10K+ GPU cluster, 2024-03: Cross-region training at 30km/50km with >98% efficiency, 2025: Delivered first fully domestic end-to-end AI cluster, Launched AI computing switch, 1500km distributed training with China Unicom, 2026-04: Filed for Hong Kong IPO (HKEX), D轮 valuation reached ¥9.16B (~$1.26B)

Growth metrics

Revenue: ¥0.318B (2023) → ¥3.247B (2024) → ¥5.203B (2025), CAGR 304.5%. Profitable since 2024 (adj. net profit ¥0.203B in 2024, ¥0.311B in 2025). Cumulative 90,000+ GPUs delivered, 34,000+ PFLOPS managed. 48 customers in 2025.

Market positioning

China's largest independent AI computing cluster provider. Ranked 9th overall among all AI cluster providers in China (1.1% market share by 2025 revenue), but #1 among independent (non-cloud-giant-affiliated) providers.

Geographic focus

China (domestic market primarily, with stated ambitions for overseas expansion)

About 胡效赫 (Hu Xiaohe)

Ex-Tsinghua University (PhD, Postdoc, Network Security Lab); UC Berkeley Visiting Scholar (Scott Shenker lab). Led development of China's first carrier-grade Tbps programmable network product. Successfully trained a 70B-parameter model on a 1000-GPU cluster on a supercomputing platform.

Official website: