SiliconFlow Raises $294 Million as China’s AI Inference Demand Surges
Chinese AI inference startup SiliconFlow has raised over 2 billion yuan ($294 million) in a Series B funding round backed by an unprecedented coalition of more than 17 corporate investors spanning energy, chip manufacturing, cloud computing, telecom, and AI sectors, according to Caixin Global. The round marks one of the largest funding events in China’s AI infrastructure space and the company’s fifth capital raise since its founding in August 2023.
Context
SiliconFlow (硅基流动) was founded by Yuan Jinhui (袁进辉), a PhD and former chief architect of the deep learning framework OneFlow, which was acquired by Meituan in 2023. The Beijing-based startup focuses exclusively on AI inference optimization — the process of running trained AI models to generate outputs — positioning itself as a critical “middle layer” between chip makers and cloud giants in China’s fragmented AI hardware ecosystem.
Key Developments
The Series B round was led by an extraordinary cross-industry investor group that includes Trip.com Group, SenseTime, Jinko Solar Holdings, Kingdee International, China Unicom Venture Investment, Nio Capital, Biren Technology, Giant Interactive Group, GGV Capital, and several state-linked entities such as China Development Bank Capital and Zhongguancun Science City, as reported by Caixin. China Renaissance acted as the exclusive financial advisor.
According to SiliconFlow’s official announcement, the company processes approximately 1 trillion tokens daily through its SiliconCloud platform, which hosts over 160 AI models. The platform has registered more than 10 million users and serves over 10,000 enterprise clients. Revenue grew tenfold year-over-year, with international revenue reaching millions of U.S. dollars per month.
SiliconFlow ranks fourth in China’s public cloud MaaS (Model-as-a-Service) market by token consumption, according to IDC’s 2025 data — the only startup in the top five behind ByteDance Volcano Engine, Alibaba Cloud, and Baidu AI Cloud. The company has also been recognized by Gartner in its “Global AI Cloud Market Guide” and “China AI Infrastructure Market Guide,” the only startup included in the latter.
Globally, SiliconFlow ranked first on OpenRouter for daily token consumption for several consecutive weeks among 70+ token suppliers, and its Dify Marketplace plugin installations exceeded 550,000, placing it in the top five among 100+ suppliers, as detailed by 36Kr.
Analysis
The investment carries strategic significance beyond its size. The involvement of state-linked investors — China Unicom, China Development Bank Capital, and Zhongguancun Science City — signals that SiliconFlow is being positioned as a national AI infrastructure asset, not merely a venture-backed startup.
Jinko Solar Holdings described the investment as a “strong alliance at both capital and industrial levels, opening up the full ‘power-computing-model-platform-application’ chain,” highlighting the convergence of energy and AI infrastructure. Biren Technology, a domestic GPU maker, noted that SiliconFlow is “the first AI infrastructure enterprise to achieve a closed loop of ‘inference deployment - computing scheduling - application landing’ on multiple domestic computing chips.”
This “full industry chain” investor coalition reflects a broader trend in China: as U.S. export controls restrict access to advanced NVIDIA chips, companies that can optimize inference across multiple domestic chip architectures — including Huawei Ascend, Biren, Muxi, and Moore Threads — are becoming strategically vital. SiliconFlow claims its technology can increase model speeds tenfold and cut costs by 46%.
What’s Next
The Chinese MaaS market is experiencing explosive growth driven by the enterprise AI agent boom. IDC projects that China’s enterprise MaaS market token consumption will jump twentyfold in 2026, generating approximately 18.6 billion yuan in revenue. Total token consumption across China is expected to reach 40,000 trillion tokens in 2026, up roughly 20x from 2025.
With this Series B funding, SiliconFlow is well-positioned to accelerate its “Token factory” model — treating AI inference tokens as a commodity akin to electricity or water. The company’s rapid escalation in funding size, from a 50 million yuan angel round in January 2024 to over 2 billion yuan now — roughly a 40x increase in 2.5 years — underscores the intensity of investor conviction in the AI inference infrastructure thesis.
Key questions remain, however. The company’s specific valuation was not disclosed, and its profitability status remains unclear. As a Chinese AI company dependent on access to advanced chips, SiliconFlow also faces ongoing risk from U.S. export controls. And with ByteDance and Alibaba Cloud commanding vastly greater resources, the question of whether an independent inference platform can maintain its competitive position against the cloud giants remains open.