Produced by | Bullet Finance

Art Design | Qianqian

Reviewed by | Songwen

As tokens are seen as the 'utilities' of the AI era, capital competition around inference computing power is rapidly intensifying. Recently, Silicon Flow submitted its prospectus to the Hong Kong Stock Exchange, aiming to become the 'first AI token factory stock.'

Founded in August 2023, this young company has completed seven rounds of financing at an astonishing pace and set a record for the largest single funding round in China's domestic third-party MaaS (Model as a Service) sector in 2026, attracting investments from Alibaba, Meituan, and others, with a post-investment valuation reaching RMB 7.74 billion.

Behind the narrative of rapid capital growth lies the harsh reality of cumulative net losses of RMB 440 million over three years and a negative gross margin of -24% in 2025, with the latter drawing significant market attention.

As an independent intermediate layer, Silicon Flow widely adapts to diverse chips such as NVIDIA and Ascend, supporting 170 mainstream models including DeepSeek. However, as China's fourth-largest token supply platform, its 1.5% market share pales in comparison to the three industry giants.

When leading industry players possess significant pricing power and ecological barriers, the survival space for third-party platforms is severely compressed. Their efforts to acquire users through subsidies may inadvertently accelerate their own financial bleeding.

Meanwhile, price fluctuations further complicate Silicon Flow's situation. Focusing on optimizing domestic computing power and private deployments for government and enterprise clients—areas that major players 'are unwilling or find uneconomical to pursue'—Silicon Flow must navigate the challenge of balancing 'avoiding giants' with 'achieving economies of scale.' This will be a core test for its IPO narrative.

Continuous capital injections provide ammunition for this breakthrough but also set a time window. As capital patience races against burning money, can its business model as an independent AI infrastructure service provider withstand profitability tests?

The so-called 'Token Factory' refers to an MaaS infrastructure platform that standardizes and packages underlying GPU/domestic chip computing power through computing power scheduling and inference engines, charging externally based on token (token element) usage volume.

According to the prospectus, Silicon Flow positions itself as an 'intermediate layer' in AI inference infrastructure, transforming underlying computing power into standardized token supply through its self-developed inference engines and computing power orchestration systems, thus earning the analogy of a 'Token Factory.'

Commercially, the company adopts two main implementation paths:

1. Private deployment services (i.e., on-premises solutions), providing software solutions for enterprises that have already purchased computing power to upgrade their internal computing resources into a token factory with a single click;

2. Public cloud services, offering out-of-the-box token services to technology-oriented enterprises.

In 2025, public cloud services accounted for 52.9% of revenue, while on-premises solutions contributed 47.1%, making public cloud the largest revenue source.

(Figure/Prospectus)

The prospectus reveals that the company's revenue surged from RMB 7.35 million in 2024 to RMB 55.33 million in 2025, a year-on-year increase of 653%. By April 30, 2026, the platform had surpassed 10.28 million registered users, with an average daily token throughput of approximately 578.5 billion.

This explosive growth was driven by a 1602.6% year-on-year increase in China's token supply market and the company's rapid expansion in both public cloud services and on-premises solutions.

In terms of token throughput, China's token supply market grew by 1602.6% from 2024 to 2025 and is expected to reach approximately 532 quadrillion tokens by 2030, with a compound annual growth rate of 638.3% from 2025 to 2030.

Notably, Silicon Flow is classified as a non-commercialized company under Hong Kong Stock Exchange's Chapter 18C. The company expects to meet the revenue requirements for a commercialized company under Chapter 18C by the end of 2026, i.e., achieving RMB 250 million in revenue for the most recent fiscal year. At the current exchange rate, this equates to approximately RMB 216 million, meaning Silicon Flow's revenue growth rate in 2026 must reach at least approximately 293%.

However, this industry-wide explosive growth has not translated into improved profitability.

In 2023 (four months from inception to year-end), 2024, and 2025, Silicon Flow incurred losses of RMB 12.22 million, RMB 81.92 million, and RMB 345 million, respectively, with cumulative losses over three years nearing RMB 440 million. Even after excluding non-cash items such as share-based payments, the adjusted net loss in 2025 still reached RMB 187 million.

(Figure/Prospectus)

A sharp decline in gross margins more intuitive ly reveals the essence of the problem. The company's overall gross margin fell from 83.3% in 2023 to 39.4% in 2024 and further turned negative to -24% in 2025.

Notably, the gross margin for the public cloud services segment plummeted to -119%, severely dragging down the overall gross margin. Even though the gross margin for on-premises solutions reached 82.5%, it could not reverse the trend.

(Figure/Prospectus)

The overall profit margin contraction primarily stems from substantial costs incurred in leasing computing power. The prospectus discloses that Silicon Flow's sales and marketing expenses skyrocketed by 1210% from RMB 6.39 million in 2024 to RMB 83.74 million in 2025, accounting for 151.4% of revenue.

Of this, RMB 54.213 million (64.7%) was spent on distributing free token vouchers, essentially using subsidies to acquire users.

Additionally, high research and development (R&D) expenditures further exacerbated cash flow pressures. In 2025, the company's R&D expenses reached RMB 209 million, 3.78 times its annual revenue, primarily used for iterating its self-developed inference engines and computing power orchestration systems.

Although the company states that continued investment in R&D is to support long-term growth and scale expansion, the combination of high R&D expenditures and a business structure with negative gross margins means that losses are unlikely to be reversed in the short term.

Regarding issues such as negative gross margins, Silicon Flow told Bullet Finance that its current overall positive gross margins primarily come from dedicated instances and on-premises businesses. The main losses stem from the public cloud Serverless API business for small and medium-sized users, which is currently in a stage of strategic losses. As losses in the Serverless API business continue to narrow, the company as a whole can achieve break-even.

Currently, the token supply market mainly forms two types of business models: independent ecosystems and closed ecosystems.

In an independent ecosystem, token supply platforms typically do not bind to any cloud provider, model, or application scenario. Instead, they provide multi-model, multi-computing power, and cross-deployment environment token supply services by connecting various types of computing power, diverse models, different computing centers, and enterprise customer needs.

In a closed ecosystem, cloud service providers, large technology companies, or model firms leverage their proprietary cloud resources, models, computing power infrastructure, and application scenarios to achieve resource integration, model invocation, and service closure within their own ecosystems.

Silicon Flow positions itself as an 'independent ecosystem token supply platform,' not binding to any single cloud, chip, or model vendor. Instead, it serves as a system software intermediate layer connecting upstream computing power, midstream models, and downstream applications.

According to the prospectus, the company's platform supports a cumulative total of 170 models, including mainstream advanced models such as DeepSeek, GLM, Kimi, and MiniMax. It ranks first globally in cross-chip and multi-model adaptability, supporting internationally leading chips (e.g., NVIDIA and AMD GPUs) as well as major domestic AI chips (including Huawei Ascend, Moore Threads, and others).

(Figure/Prospectus)

According to Frost & Sullivan data, based on token throughput in 2025, Silicon Flow is China's largest independent ecosystem token supplier, with a market share of 1.5%.

(Figure/Prospectus)

However, when including closed ecosystem vendors that rely on proprietary cloud resources and models, Silicon Flow ranks only as China's fourth-largest token supply platform, with a market share gap from the top three that is not on the same scale.

An IDC report shows that in the first half of 2025, Volcano Engine led with a 49.2% market share, followed by Alibaba Cloud at 27% and Baidu Intelligent Cloud at 17%. Together, these three captured approximately 93% of the market. (Editor's note: This data statistical caliber is limited to the public cloud MaaS market.)

In Silicon Flow's view, this position holds unique value: closed ecosystems tend to lock customers into their own computing power and models, while Silicon Flow maintains neutrality and openness. In China's diversified domestic computing power landscape with multiple competitors, this neutral positioning is more likely to earn trust, making Silicon Flow a hub connecting chip vendors, model providers, and enterprise customers.

However, customer concentration is a hidden concern. In 2023, 2024, and 2025, the top five customers accounted for 100%, 85%, and 45% of revenue, respectively, with the largest single customer's share dropping from 83.3% to 61.1% and then to 13.6%.

Although concentration is declining, the top two customers still accounted for 14% and 13% of revenue in 2025. In an industry environment where token pricing continues to decline, large customers' bargaining power is on the rise, limiting the company's ability to pass costs upstream.

More challenges arise from price wars. According to China Academy of Information and Communications Technology data, the average price of domestic large model APIs has cumulatively dropped by over 90% compared to 2023, with leading large model vendors reducing API prices more than ten times.

Represented by DeepSeek, in June 2026, DeepSeek V4 Pro officially launched a permanent price reduction, with the adjusted price Only for one-fourth of the original.

As an intermediate layer, Silicon Flow cannot control upstream computing power costs while facing continuous downward pricing pressure downstream, squeezing its profit margins from both directions.

Regarding market competition strategies, Silicon Flow's senior management publicly stated last year that the company focuses on areas where major players currently show weak interest or find uneconomical, particularly optimizing domestic computing power and serving cost-sensitive medium-to-large enterprises and government clients requiring private deployments. It aims to penetrate more enterprises with scaling application needs beyond top-tier clients.

Deploying in areas that major players avoid is a rational choice to secure survival space. However, whether this differentiated path of avoiding giant competition can prevent the company from being locked into a 'high-input, low-margin, slow-growth' trap in the long term remains to be seen.

Regarding these issues, Silicon Flow told Bullet Finance that in the field of domestic chip adaptation, it provides rich computing power solutions and is the only heterogeneous computing power platform adaptation ing NVIDIA + mainstream domestic computing power, meeting diverse downstream customer needs. Its on-premises solutions have deeply penetrated high-value-added sectors such as the internet, finance, energy, and transportation, with landmark customer cases implemented. This business primarily provides standardized software solutions with positive gross margins, avoiding 'diseconomies of scale.'

Leveraging its neutral positioning, Silicon Flow has forged a unique path in strategic collaboration, securing capital investments from across the internet, chip, computing power, energy, and telecommunications sectors.

In terms of financing pace, Silicon Flow has been highly favored by capital, completing seven rounds of financing since launching its Angel Round in December 2023.

On June 16, 2026, the company officially announced the completion of a Series B funding round exceeding RMB 2 billion, setting a record for the largest single funding round in China's domestic third-party MaaS sector in 2026. This brought its post-investment valuation to RMB 7.74 billion.

However, Bullet Finance noted that Silicon Flow's prospectus submitted to the Hong Kong Stock Exchange on June 30 shows Series B and Series B+ funding amounts of RMB 520 million and RMB 740 million, respectively, totaling RMB 1.26 billion in raised capital.

(Figure/Silicon Flow Prospectus)

In fact, Hong Kong Stock Exchange's Chapter 18C explicitly requires non-commercialized companies to have a market capitalization of no less than HKD 8 billion. After announcing the 'completion of a Series B funding round exceeding RMB 2 billion,' Silicon Flow quickly raised its market-perceived valuation to RMB 7.74 billion (approximately HKD 8.9 billion), precisely meeting the listing requirements.

Notably, Silicon Flow's prospectus discloses that its post-investment valuation after the Series B funding on April 30, 2026, was RMB 5.02 billion, falling short of listing requirements. However, after the Series B+ funding in early June, its valuation rapidly rose to RMB 7.74 billion, just meeting the market cap requirement, and it promptly submitted its application to the Hong Kong Stock Exchange on June 30. This operational path has sparked external discussions.

The prospectus reveals that Silicon Flow's shareholder list reads like a 'dream team': industry giants such as Alibaba, Huawei Hubble, Zhipu, Meituan, SenseTime, Ctrip, Jinko Solar, China Unicom, Biren, NIO Capital, and Kingdee, along with top financial institutions and state-owned capital firms like GGV Capital, Huakong Fund, and China Development Financial.

This full-industry-chain investment brings not only capital but also deep ecological synergies in scenarios, computing power, models, and markets.

For example, Biren Technology has expressed expectations for deep cooperation with Silicon Flow in chip adaptation, inference acceleration optimization, and large-scale deployment of computing power clusters to jointly build high-performance token factories.

Additionally, in February 2025, Silicon Flow was the first in the industry to launch a full-scale DeepSeek R1/V3 model based on Huawei Ascend cloud computing power, becoming the first company to successfully deploy DeepSeek on domestic chips. This event quickly garnered industry attention and found a key promoter for Huawei's Ascend ecosystem.

However, this neutral model also faces challenges. Some investors are potential competitors. Alibaba Cloud, SenseTime, and Zhipu, all shareholders, are also engaged in MaaS businesses, with differing interest demands that may affect long-term strategic decisions. For instance, Alibaba has its own Alibaba Cloud Bailian, SenseTime has its self-developed 'Wanxiang' infrastructure, and Zhipu has its own API business, creating direct competition with Silicon Flow.

Another noteworthy detail is the overlap between customers and suppliers.

Customer I (also Supplier J) was Silicon Base Flow's second-largest customer in the second quarter of 2025 and its fifth-largest supplier in the same year. Both belong to the same group. The company primarily provided AI model chip adaptation services to them, while they, in turn, mainly supplied the company with computing resources through different entities within the group.

In 2025, Silicon Base Flow's sales to Customer I amounted to RMB 7.1 million, accounting for 12.8% of the company's total revenue for the year. During the same period, the company's purchases from Supplier J totaled RMB 12.6 million, representing 7.6% of its total procurement.

(Figure / Prospectus)

Such a relationship where an investor is "both a major customer and a major supplier" is not uncommon in business, but it often implies the company's multiple dependencies on the same entity in its operations.

Regarding issues such as the company's operational independence, Silicon Base Flow stated to "Bullet Finance" that the company has autonomy in its daily operational decisions and maintains a complementary and win-win ecological synergy with industrial investors, focusing on growing the market together. Silicon Base Flow's open and neutral positioning can be sustained in the long term.

From the current perspective, Silicon Base Flow's future hinges on a practical question: Can economies of scale outpace the high costs of research and development and computing power?

Against the backdrop of tokens becoming the "utilities" of the AI era, Silicon Base Flow, as an independent infrastructure provider, has the potential to occupy a key ecological niche. However, this The premise (premise, translated to maintain context) is that it must survive the current cash-burning phase and defend its market share amid competition from industry giants.

The neutrality that Silicon Base Flow claims to uphold ultimately needs to be validated through a sustainable unit economic model, rather than merely remaining at the narrative level of ecological positioning.

*The featured image in the article is sourced from: Shetuwang, based on the VRF protocol.