SiliconFlow Bets on HKEx Listing to Fund China's AI Token Infrastructure Race
China's first dedicated AI token supply platform is heading to public markets, forcing investors to weigh a 653% revenue surge against a gross margin that has turned deeply negative — a tension that defines the entire AI infrastructure buildout cycle.
Beijing SiliconFlow Technolog filed a listing application with the Hong Kong Stock Exchange on July 8, 2026, seeking a main board debut under Chapter 18C — the bourse's specialist technology company framework designed for pre-profit innovators. Huatai International and Haitong International serve as joint sponsors. The filing positions SiliconFlow as what Frost & Sullivan characterizes as China's leading independent-ecosystem token supply platform, a niche that sits squarely in the critical middle layer between raw GPU compute and end-user AI applications.
Revenue Scaling Rapidly, But Unit Economics Turn Negative
SiliconFlow's top-line trajectory is difficult to ignore: revenue reached RMB 55.33 million (approximately US$7.7 million) in 2025, a 653% year-on-year jump from RMB 7.35 million in 2024. The company recorded a nominal RMB 6,000 in revenue during its partial first year of operation ending December 31, 2023.
Yet the more consequential number for institutional investors is the gross margin collapse. SiliconFlow's gross margin fell from 39.4% in 2024 to negative 24% in 2025 — meaning the company is currently losing RMB 0.24 for every RMB 1.00 of token service it delivers. The proximate cause is a deliberate land-grab strategy: to secure market share and sustain throughput growth, the company has been purchasing compute capacity at a pace that materially outstrips revenue, while simultaneously issuing free token vouchers to acquire users.
Sales and marketing expenditure surged 1,210% year-on-year to RMB 83.74 million in 2025, with promotional compute resource costs — effectively free token credits — accounting for 64.7% of that line. Research and development spending reached RMB 209 million in 2025, equivalent to 3.78 times full-year revenue. The resulting net losses have compounded sharply: RMB 12.22 million in 2023, RMB 81.92 million in 2024, and RMB 345 million in 2025. Cumulative losses across the reporting period stand at approximately RMB 450 million (US$62.5 million).
Public Cloud Shift Signals a Structural Business Model Transition
Buried within the loss figures is a revenue-mix pivot that carries strategic significance. In 2024, on-premises deployment solutions dominated SiliconFlow's revenue at 85.4%, with public cloud services contributing just 14.6%. By 2025, public cloud revenue had grown to RMB 29.26 million, representing 52.9% of total revenue — crossing the majority threshold for the first time.
This inversion matters for two reasons. First, public cloud token services carry a fundamentally different unit economics profile over time: higher initial subsidy costs but stronger margin leverage at scale, as compute utilization rates improve. Second, the shift toward cloud-delivered tokens aligns with how enterprise AI adoption is evolving in China — developers increasingly prefer API-based, pay-per-token consumption over capital-intensive on-premises GPU clusters.
As of April 30, 2026, SiliconFlow's platform had registered more than 10 million users and served over 13,000 enterprise clients, with average daily token throughput of approximately 578.5 billion tokens in April 2026 and a single-day peak of roughly 1.07 trillion tokens. The platform supports more than 170 mainstream AI models. Frost & Sullivan ranked SiliconFlow as China's fourth-largest token supply platform by annual token throughput in 2025, with a 1.5% market share.
Heavyweight Backers Validate the Infrastructure Thesis
SiliconFlow has completed seven funding rounds in under three years — a cadence that reflects both the urgency of the AI infrastructure race and the credibility of its founding team. Valuation has climbed from RMB 280 million (US$38.9 million) at the angel round to RMB 7.74 billion (US$1.075 billion) at the Series B+ round, crossing unicorn status.
The investor roster reads as a cross-section of China's technology establishment: Alibaba, Huawei, Meituan, Zhipu AI, Trip.com Group, Biren Technology, NIO Capital, and SenseTime all hold positions. Alibaba, through multiple Hangzhou-registered entities, holds 7.42% of shares, making it the largest external institutional shareholder. Huawei's investment arm Hubble Technology holds 4.07%, and Beijing Sinovation Ventures holds 4.01%.
Founder and CEO Yuan Jinhui retains 14.35% direct equity and controls 44.48% of voting rights in aggregate with four employee incentive platforms — a structure designed to preserve founder control post-listing. Yuan holds a PhD from Tsinghua University, where he studied under Zhang Bo, a Chinese Academy of Sciences academician widely regarded as a founding figure of Chinese AI research. He previously led the development of the LightLDA high-efficiency topic model training algorithm at Microsoft China and founded the OneFlow deep learning framework before pivoting to SiliconFlow in late 2023, following the acquisition of his prior venture by Meituan.
Assessing the Investment Case: Infrastructure Premium vs. Burn Rate Risk
SiliconFlow's IPO arrives at an inflection point for China's AI stack. With large model development increasingly commoditized — dozens of foundation models now compete on near-identical benchmarks — the competitive moat is shifting downstream toward inference efficiency, compute orchestration, and developer ecosystem lock-in. SiliconFlow's proprietary inference engine and heterogeneous compute scheduling system, which aggregates GPU resources across multiple vendors and architectures into standardized token services, positions it as a picks-and-shovels play on AI adoption broadly rather than a bet on any single model.
The bear case centers on sustainability. A gross margin of negative 24% is not a temporary dip — it reflects a market where token pricing has been driven to sub-cost levels by competitive subsidization, including from hyperscalers with far deeper balance sheets. SiliconFlow's RMB 450 million cumulative loss on under RMB 70 million in cumulative revenue implies that the current business model requires either a significant pricing recovery, a step-change in compute efficiency, or continued external capital to remain viable. The IPO proceeds are therefore less a validation of maturity than a necessary fuel injection for the next phase of the land-grab.
Chapter 18C was specifically engineered for this profile of company, and Hong Kong's market has demonstrated appetite for pre-profit AI names. Whether SiliconFlow can convert its throughput scale and blue-chip investor base into a credible path to positive unit economics will be the central question facing underwriters as the roadshow approaches.
Related Coverage:
Understanding China's AI Ecosystem: Five Forces Shaping Competition and Growth