SiliconFlow
Category: AI Infrastructure
China's leading independent AI token supply platform and AI infrastructure provider, offering high-performance, cost-effective model inference services via its SiliconCloud MaaS platform. SiliconFlow was founded in 2023. The company is led by 袁进辉 (Jinhui Yuan). Based in Beijing, China. Team size: 101-500. Total funding raised: ~$275M (¥1.87B across 7 rounds). Latest round: Series B. Key investors include 阿里巴巴 (Alibaba), 华为哈勃 (Huawei Hubble), 美团 (Meituan), 创新工场 (Sinovation Ventures), 携程战投 (Trip.com Group), 耀途资本 (Yao Capital), 华创资本 (China Growth Capital), 普华资本 (Puhua Capital), 奇绩创坛 (MiraclePlus), 智谱AI (Zhipu AI), 360, 王慧文 (Wang Huiwen).
AMW Analysis
SiliconFlow is a Beijing-based AI infrastructure provider that operates an independent token supply platform, offering model inference services through its SiliconCloud MaaS platform. It aggregates heterogeneous compute from NVIDIA, AMD, and Huawei Ascend, using a proprietary inference engine to deliver standardized, low-cost token supply for LLMs and multimodal models. Founded in 2023, the company has raised approximately $275 million across seven rounds, with strategic backing from Alibaba, Huawei Hubble, Meituan, and others.
Recent news shows SiliconFlow has filed for a Hong Kong IPO under Chapter 18C special-tech listing rules, positioning itself as the city's first publicly listed "AI Token Factory" stock. Reports indicate the filing targets a valuation of $1.07 billion, though another source cites a $7.7 billion valuation, and note mounting losses. The company completed seven funding rounds in under three years, reflecting rapid capital accumulation and investor interest from major Chinese tech firms. The IPO filing marks a significant step in its trajectory from a startup to a public market candidate, though financial sustainability remains a point of attention.
AMW analysis, generated from 2 tracked news signals.
- Founded
- 2023
- Headquarters
- Beijing, China
- Team size
- 101-500
- Total funding
- ~$275M (¥1.87B across 7 rounds)
Value proposition
Neutral, independent AI infrastructure platform that aggregates heterogeneous compute (NVIDIA, AMD, Huawei Ascend, etc.) with a proprietary inference engine to deliver standardized, high-speed, low-cost token supply for LLMs and multimodal models, enabling developers and enterprises to deploy AI without infrastructure complexity.
Products and solutions
SiliconCloud (one-stop MaaS platform with API services for 170+ models), Reserved Instances (dedicated compute for enterprise inference), Enterprise MaaS Platform (private deployment, fine-tuning, heterogeneous compute management), ComfyUI Cloud Plugin, AI Gateway, BYOK on OpenRouter
Unique value
China's largest independent (non-hyperscaler) token supply platform with a neutral position, proprietary inference acceleration engine delivering 10x+ speed improvements, and support for 170+ open-source and commercial models via a single API — all while maintaining independence from any single cloud ecosystem.
Target customer
AI developers, startups, large internet companies, traditional enterprises, research institutions, government agencies, and AI hardware manufacturers
Industries served
Internet, Education, Government/Public Sector, Smart Computing Centers, AI Hardware, Finance, Manufacturing, Entertainment/Media
Technology advantage
Proprietary high-performance inference engine delivering 10x+ speed improvement for language models and 3x+ for image generation; heterogeneous compute orchestration across NVIDIA, AMD, Huawei Ascend, and domestic Chinese chips; cost optimization achieving 46-66% cost savings; self-developed model inference acceleration and compute scheduling systems
How they differentiate
Positioned as a neutral, independent token supplier — not tied to any single hyperscaler cloud — enabling multi-cloud and cross-ecosystem deployment. First to deploy DeepSeek R1/V3 on domestic Huawei Ascend chips. Aggressive pricing and broadest model selection among independent Chinese MaaS platforms. Serves as the #1 third-party model provider by daily token usage on OpenRouter globally.
Main competitors
火山引擎 (Volcengine/ByteDance) — #1 by market share, 阿里云百炼 (Alibaba Cloud Bailian), 百度智能云 (Baidu AI Cloud), 华为云 (Huawei Cloud), 腾讯云 (Tencent Cloud), 潞晨科技 (Luchen Tech), 七牛云 AI (Qiniu Cloud AI), OpenRouter (global), Fireworks AI (US)
Key partnerships
阿里巴巴 (Alibaba) — strategic investor & cloud partner, 华为哈勃 (Huawei Hubble) — investor, joint DeepSeek R1/V3 on Ascend, 美团 (Meituan) — strategic investor, 携程 (Trip.com Group) — strategic investor, 智谱AI (Zhipu AI) — investor & partner, 360 — investor, 晶科能源 (Jinko Energy) — investor, 金蝶 (Kingdee) — investor, 中国联通 (China Unicom Xinwo) — investor, 壁仞科技 (BirenTech) — investor & chip partner, 商汤科技 (SenseTime) — investor & partner, 蔚来资本 (NIO Capital) — investor
Notable customers
13,000+ enterprise customers including government agencies, internet companies, financial institutions, and manufacturing firms, 10.28M+ registered users (as of Apr 2026), 716,000+ paying customers for serverless token service (2025), 49 dedicated instance enterprise customers (2025), 20 local deployment customers (2025)
Major milestones
2023-08: Company founded, 2023-12: ¥47.2M Angel round (奇绩创坛,水木清华,创新工场,耀途), 2024-05: SiliconCloud platform launched, 2024-05~2025-01: Angel+ round (华为哈勃入局), 2025-02: First to deploy DeepSeek R1/V3 on Huawei Ascend cloud, 2025-02: ¥71.48M Pre-A round led by Alibaba, 2025-05~06: ¥286M Series A (美团,360,商汤,壁仞进入), 2025-12: 10,000+ enterprise customers, 150+ models, 2026-02~03: ¥220M A+ round, 2026-04: 10M+ registered users, 2026-06-05: ¥520M Series B, 2026-06-17: ¥740M B+ round (valuation ¥7.74B)
Growth metrics
Revenue: ¥55.33M (2025), up 653% YoY from ¥7.35M (2024); Registered users: 10.28M+ (as of Apr 2026); Enterprise customers: 13,000+; Daily token throughput: 5.785T (Apr 2026 avg), peak 1.07T single day; 170+ models on platform; 2025 net loss: ¥345M; 2025 gross margin: -24%; 2025 R&D: ¥209M; 2025 cumulative net loss: ~¥440M (2023-2025); Cash & equivalents: ¥172M (end 2025)
Market positioning
China's largest independent (non-hyperscaler) token supply platform with 1.5% market share; #4 among all token suppliers in China (including hyperscalers) by 2025 annual token throughput; #1 third-party model provider by daily token usage on OpenRouter globally
Geographic focus
China (primary), with international expansion via siliconflow.com and OpenRouter integration
Patents and IP
Proprietary inference acceleration engine; heterogeneous compute orchestration system; model deployment and optimization IP inherited from OneFlow deep learning framework team
About 袁进辉 (Jinhui Yuan)
Ex-Founder & CEO of OneFlow (一流科技); Ex-Principal Researcher at Microsoft Research Asia (2013-2016), invented LightLDA (world's fastest large-scale topic model training system); PhD in Computer Science from Tsinghua University
Latest news about SiliconFlow
More AI Infrastructure companies
Official website: https://siliconflow.cn
