Observed Signal · Jul 6, 2026 · IPO Filing · Source: Hello China Tech · Impact: 3/5 · Sentiment: Negative
SiliconFlow IPO Reveals Cost of China's Token Boom
SiliconFlow, founded in August 2023, is China’s largest independent token-supply platform and on June 30 filed for a Hong Kong listing under Chapter 18C. As of April 2026 it reported processing an average 578.5 billion tokens per day across 170+ models, with more than 10 million registered users and over 13,000 enterprise customers, but holds only ~1.5% market share. The prospectus shows rapid revenue growth (Rmb 55.3m, +653%) alongside dire unit economics: public cloud services (52.9% of 2025 revenue) had a gross margin of negative 119% in 2025, pulled down blended margins and drove a net loss of Rmb 345.5m. The filing highlights heavy compute rental costs, steep market price cuts from major suppliers, and reliance on promotional compute credits to acquire developers, raising questions about the commercial viability of independent token intermediaries.
The Hong Kong IPO prospectus provides a rare, audited view into token-supply economics and cloud-dependent unit economics, informing competition, pricing dynamics, and the viability of independent inference intermediaries.
Track ByteDance Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- SiliconFlow was founded in August 2023 and filed for a Hong Kong listing under Chapter 18C on June 30, 2026.
- As of April 2026 SiliconFlow processed an average of 578.5 billion tokens per day across more than 170 models, serving over 10 million registered users and 13,000+ enterprise customers.
- SiliconFlow reported revenue of Rmb 55.3 million in 2025 (a 653% increase) but a net loss of Rmb 345.5 million; adjusted loss (excluding share-based compensation) was Rmb 187 million.
- Public cloud services made up 52.9% of 2025 revenue but had a gross margin of negative 119% in 2025; compute rental accounted for 86.9% of cost of goods sold.
- Major model/API price cuts in May 2026 included DeepSeek (75%), Xiaomi (up to 99%), and Tencent Cloud (up to 97.5%); promotional compute credits comprised over 64% of the company’s Rmb 83.7m selling and marketing expense.
Connected Companies & Entities
4 Entities mapped“The top three token suppliers by throughput, Volcengine (ByteDance’s cloud arm), Alibaba Cloud, and Baidu AI Cloud, are all hyperscaler divi...”
“The top three token suppliers by throughput, Volcengine (ByteDance’s cloud arm), Alibaba Cloud, and Baidu AI Cloud, are all hyperscaler divi...”
“In May 2026, DeepSeek, the Hangzhou-based AI lab, made a permanent 75% cut to its V4-Pro API price....”
“Xiaomi and Tencent Cloud followed with maximum reductions of 99% and 97.5%, respectively....”
Ontology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
China's Token Factory Models Reshape Inference Economics
The article analyzes three commercial models emerging in China for selling AI inference output called 'tokens': (1) a capacity-procurement framework (China Telecom’s Ningxia unit launched bidding for 'Token generation capacity services' with a five-year, Rmb 17.4bn ceiling); (2) a hardware-focused procurement (Digital China submitted a Rmb 717m bid for a 'Domestic AI Computing Token Factory' covering super-node servers); and (3) a managed operating model (Approaching AI reportedly raised >Rmb 1bn and, with a fund linked to Henan Investment Group, plans a Token Factory to run inference without owning chips). The piece also notes SiliconFlow’s Hong Kong IPO filing showing deeply negative public cloud margins and that the China Academy of Information and Communications Technology launched a Token Service Evaluation System in 2026. The article examines whether token capacity can be contracted, financed, and operated independently of physical chips.
China Tech Newsletter Year-One Scorecard
Hello China Tech published a self-assessment scoring a year of its paid coverage against public filings and outcomes. The essay grades four specific calls: Moonshot AI's $500m round and later Hong Kong listing plans; SiliconFlow’s Hong Kong IPO filing confirming massive token volumes but weak cloud margins and low ARPU; Unitree’s IPO disclosures showing most humanoid revenue coming from research institutions; and CATL’s strategic investments expanding from batteries into data center energy and related assets. The piece identifies four testable questions for the next year — around robot IPO disclosures, open-weight models converting usage to paid workloads outside China, AI supply-chain splits and re-fusions amid export controls, and what China’s application-layer products teach global markets — and notes two previously paid essays temporarily open to all readers.
China's AI Enters a Perpetual Capital Cycle
China’s AI industry is experiencing rapid, recurrent fundraising where IPOs serve as channels for continuous capital injections rather than endpoints. In early July 2026 several listed and private AI companies returned to the market: Biren Technology raised $900m, Iluvatar CoreX raised $902m, Zhipu priced a $4bn share sale, DeepSeek discussed a new round at a $71bn pre-money valuation after a prior >$7bn round, and CXMT priced an IPO that could raise up to $9.8bn. Prospectuses filed under Hong Kong’s Chapter 18C reveal structural gaps between revenue and R&D/spending (e.g., Biren’s 2024 R&D was ~2.5× revenue), meaning continued external capital is required to finance commercialization, fabrication, component procurement, data centers, and scaling.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.