China's largest independent ecological token supplier, SiliconFlow, files for listing on the Hong Kong Stock Exchange.

According to the Hong Kong Stock Exchange disclosure on June 30, Beijing SiliconFlow Technology Co., Ltd. (referred to as "SiliconFlow") has officially submitted its main board listing application, with Huatai International and Haitong International acting as joint sponsors.

As an AI infrastructure service provider established in August 2023, SiliconFlow has targeted the capital market in less than three years. The prospectus shows that with the large-scale application of foundational models in enterprises, SiliconFlow's business volume has expanded rapidly. However, at the same time, the high initial cost of computing power leasing has led to a negative gross profit margin in its financials.

The founding of SiliconFlow has a strong background of technological continuity. The company's founder, chairman, and CEO, Dr. Yuan Jinhui, previously founded the open-source deep learning underlying framework company Beijing OneFlow Technology Co., Ltd. As the industry's focus gradually shifts from model training to inference, SiliconFlow was officially incorporated in August 2023.

In June 2026, SiliconFlow completed a critical technological base integration by entering into an asset purchase agreement with OneFlow, formally acquiring the latter's intellectual property portfolio.

According to the prospectus, this move aims to enhance the company's technical capabilities in decentralized, data-flow-based inference architectures and large-scale context memory systems, further solidifying its positioning as an inference infrastructure provider. In the same month, the company was restructured as a joint-stock company, completing the main structural adjustments before its Hong Kong listing.

In terms of business model, SiliconFlow defines itself as an open, independent token supply platform. Facing the enormous token demand driven by the evolution of AI large models toward multimodal interaction and complex task execution, SiliconFlow does not directly develop user-facing AI application terminals. Instead, it connects underlying heterogeneous computing resources—such as chips with different architectures—with various AI large models through a system software layer, delivering standardized token services to developers and enterprises.

Specifically, its core business is divided into two major segments: public cloud services and on-premises deployment solutions.

Public cloud services include pay-as-you-go serverless token services and dedicated instances that reserve computing power for enterprises with higher performance and stability requirements. On-premises deployment involves directly deploying the inference engine and computing resource orchestration system into the data centers of large clients.

In terms of market position, Frost & Sullivan data shows that based on annual token throughput in 2025, SiliconFlow held a 1.5% share of the Chinese token supply market, ranking fourth, and ranked first among all independent ecosystem token suppliers.

In terms of operating data, its growth curve is relatively steep. As of April 30, 2026, the platform's registered users had exceeded 10 million. In April 2026, the average daily token throughput reached 578.5 billion, with a single-day record of 1.0714 trillion. As of the latest practicable date of the prospectus, the platform had supported over 170 models in total and served over 13,000 enterprise customers.

The heavy reliance on computing power for large model inference prompted SiliconFlow to initiate intensive fundraising shortly after its establishment. From December 2023 to June 2026, the company completed seven rounds of investment.

The composition of its investor base reflects the strategic positioning of industry players along the upstream and downstream chains regarding inference infrastructure.

According to the equity structure, Alibaba Group holds 7.42% through Hangzhou Various Shareholdings, making it the largest institutional shareholder. Other core venture capital institutions include Yao Tu Capital, Puhua Capital, and Zhipu AI.

Additionally, industrial capital involvement is particularly notable, including Huawei Hubble, Meituan, 360, Zhipu AI, SenseTime, and computing chip manufacturer Biren Technology.

Beyond internet and AI companies, clean energy company JinkoSolar, software service provider iSoftStone, gaming company Giant Network, and China Unicom also participated in strategic investments.

This cross-sector equity linkage objectively facilitates its computing power procurement, model integration, and commercialization in downstream application scenarios.

The financial statements objectively reflect SiliconFlow's current commercialization progress and profitability bottlenecks. In the early stage of commercialization, the company's operating data shows significant scale expansion characteristics, but profitability is directly squeezed by underlying computing power costs.

On the revenue side, the company's total revenue increased from RMB 7.346 million in 2024 to RMB 55.33 million in 2025, a year-on-year increase of 653.2%. This was primarily driven by the expansion of the customer base for public cloud and on-premises deployment solutions, with public cloud services accounting for 52.9% of revenue in 2025. In overseas markets, with the advancement of globalization strategies, its monthly overseas revenue exceeded USD 1 million in 2026.

However, the multiple-fold revenue growth did not lead to a corresponding accumulation of gross profit. The prospectus shows that in 2024, the company's gross profit was RMB 2.894 million, with a gross profit margin of 39.4%; but in 2025, the company recorded a gross loss of RMB 13.302 million, with a gross profit margin dropping to -24.0%.

The core reason for the negative gross profit margin is the surge in sales costs.

The prospectus shows that the company's sales cost soared from RMB 4.452 million in 2024 to RMB 68.632 million in 2025. This was mainly because during the rapid expansion of public cloud services, the company needed to lease a large amount of underlying computing power resources in advance to support the surging token demand, while the utilization rate of computing resources was still ramping up, causing cost growth to significantly outpace revenue growth.

In terms of profit performance, affected by high computing power leasing costs and sustained R&D investment, the company is currently in a loss state. In 2023 (August to December), 2024, and 2025, SiliconFlow's net losses were RMB 12.223 million, RMB 81.915 million, and RMB 345 million, respectively.

Among these, as a technology-driven enterprise, R&D expenditure is its largest expense. In 2025, R&D expenditure reached RMB 209 million, accounting for 378.1% of total revenue for that year. Excluding non-cash items such as share-based compensation expenses and redemption liability interest expenses, the adjusted net losses for 2024 and 2025 were RMB 54.01 million and RMB 187 million, respectively.

Regarding the cash flow position, the upfront business expansion has led to an increase in the company's cash consumption.

In 2025, the net cash used in operating activities was RMB -172 million. However, thanks to support from multiple rounds of financing, as of December 31, 2025, the company held cash and cash equivalents of RMB 172 million and time deposits of RMB 100 million.

Additionally, the company had a net liability balance of RMB 389 million on its balance sheet, but this was mainly due to redemption liabilities totaling approximately RMB 499 million issued to pre-IPO investors. After listing, these special rights will terminate, and this liability will be converted to equity, which is expected to reverse its net liability position.

Overall, SiliconFlow has entered the inference segment of the generative AI industry, which has a relatively clear commercialization path, and has achieved significant business scale and market share in its early stages. However, against the backdrop of the industry's hard physical constraints on the computing power supply chain, its financial data intuitively reflects the squeeze on profit margins for infrastructure service providers caused by underlying computing power costs.

After listing, how to further improve token output efficiency per unit of computing power through deep optimization of the inference engine, and gradually amortize the high upfront computing power leasing costs as customer stickiness increases, will be key for SiliconFlow to demonstrate the long-term sustainability of its business model to the capital market.

Risk Warning and Disclaimer

        Market risks exist, and investment should be made with caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this content is at your own risk.