SiliconFlow, the AI 'Token Factory,' Files for Hong Kong IPO After 34 Months

There’s a Chinese AI startup you may not have heard of, and it just filed for a public offering in Hong Kong. SiliconFlow — founded just 34 months ago, in August 2023 — submitted its prospectus to the Hong Kong Stock Exchange on June 30, seeking to list under Chapter 18C, the exchange’s special rule for tech companies.

The numbers in its filing tell a striking story. Revenue jumped from 7.35 million yuan in 2024 to 55.33 million yuan in 2025 — a 653% year-over-year increase. But like most AI infrastructure plays, profitability remains a work in progress. The company posted net losses of 81.92 million yuan in 2024 and 345 million yuan in 2025, with adjusted net losses of 54 million yuan and 187 million yuan respectively.

SiliconFlow describes itself as an AI capability provider, but its business is best understood as a “token factory.” The company operates a distributed inference infrastructure that processes AI model queries at massive scale — think API calls for large language models, image generation, and video generation workloads. Its daily token throughput went from 47.8 billion at the end of 2024 to 578.5 billion by April 2026, with peak throughput crossing the trillion-token mark. The company now serves over 13,000 enterprise clients and has registered more than 10 million users.

Its product lineup spans API services, dedicated compute instances, an enterprise MaaS (Model as a Service) platform, and all-in-one AI appliances for on-premise deployment. The company has built integrations across LLMs, text-to-image models, and video generation — covering much of the generative AI stack that Chinese developers and enterprises are racing to adopt.

The timing of the IPO is notable. SiliconFlow is going public under Hong Kong’s Chapter 18C rules, a framework designed specifically for pre-commercial or early-stage specialty tech companies that may not meet standard exchange financial thresholds. The rules were introduced in 2023 to attract precisely the kind of company SiliconFlow represents: high-growth, capital-intensive, and still burning cash at scale.

SiliconFlow’s growth mirrors the broader Chinese AI boom. As domestic alternatives to Western foundation models multiply, the companies providing the infrastructure to run those models — inference engines, API gateways, latency optimization layers — have become indispensable. SiliconFlow is one of several Chinese inference startups that have ridden this wave, competing with players like Zhipu AI’s API platform and ByteDance’s Volcano Engine.

Whether SiliconFlow can narrow its losses fast enough to satisfy public market investors remains an open question. A 653% revenue growth rate is impressive from any angle, but 345 million yuan in annual losses on 55 million yuan in revenue means the burn rate demands attention. The prospectus will give analysts their first real look at the unit economics of the Chinese AI inference business — and whether it’s a viable standalone model or a feature of something larger.