Can the 'AI + devices' path succeed?

《Investor China》Jiang Ji

As the large model sector transitions from a 'hundred-model battle' to an 'ecosystem positioning' deep zone, StepFun's Hong Kong IPO process is becoming a critical case study for observing the viability of AI commercialization in China. Founded just three years ago, the company is accelerating the dismantling of its offshore structure and preparing to list in Hong Kong, aiming to raise approximately $500 million with a market-rumored valuation targeting $10 billion.

However, behind the glossy capital narrative lie an unproven profit model, a channel structure highly dependent on a few hardware manufacturers, and pressure from cost re-evaluation driven by technological paradigm shifts. With Zhipu (02513.HK) and MiniMax already leading the way in Hong Kong listings, can StepFun cross the breakeven line with its differentiated 'AI + devices' approach?

From 'Tech Geek' to 'Capital Darling': StepFun's Three-Year Transformation

Founded in April 2023 by Jiang Daxin, former Microsoft Global Vice President, StepFun is headquartered in Shanghai and dedicated to artificial general intelligence (AGI) research and development. The company initially adopted a typical offshore structure in the Cayman Islands to align with the investment preferences of U.S. dollar funds. However, entering 2026, to facilitate its Hong Kong listing plan, the company is systematically dismantling its red-chip structure. While this move may increase compliance costs and extend preparation timelines, it reflects the current regulatory focus on standardizing data security and state-owned enterprise participation. In early April this year, the company formally completed its shareholding reform, transitioning from a 'limited liability company' to a 'joint-stock limited company.' Yin Qi, founder of Megvii Technology, assumed the role of chairman, signaling a clear intent to tilt the management structure toward industrial implementation and compliance governance, clearing institutional hurdles for the subsequent filing.

Financial data represents StepFun's core bargaining chip for accessing the capital market. According to *Caijing* magazine, StepFun's 2025 revenue approached RMB 500 million, with 2026 projected revenue around RMB 1.2 billion. Against the backdrop of widespread commercialization bottlenecks among AI startups, this growth rate offers some distinctiveness. The growth primarily relies on a B2B2C dual-drive model: enterprise services have covered multiple Fortune 500 companies; on the device side, its AgentAPI call volume has grown significantly. However, high revenue does not conceal the reality of strategic losses. Persistently high computational costs for large model training and inference, combined with rigid expenditures on market promotion and top talent recruitment, leave the timing of crossing the breakeven point as the most critical variable for markets ahead of the prospectus disclosure. If the prospectus fails to provide a clear path for gross margin improvement and cash flow inflection expectations, the high-valuation narrative will face fundamental scrutiny.

Hong Kong Scheduling and Valuation Negotiation: A Capital Examination with No Room for Error

StepFun's financing journey clearly reflects the shifting risk preferences of China's hard tech capital. In the early stages, dollar VCs such as Sequoia China and IDG Capital led the investments; in 2024, Series B introduced Tencent and Shanghai State-owned Investment (one of Shanghai's primary strategic investment channels for the AI industry); by January 2026, the company completed a Series B+ round exceeding RMB 5 billion, with investors including state-owned capital such as China Life Equity, Pudong Venture Capital, and Huaqin Technology, along with industrial capital, achieving a closed loop of 'VC + state-owned + industrial' shareholder structure. Entering the Pre-IPO phase, valuations have surged upward. Markets widely expect its IPO cornerstone pricing to approach $10 billion, driven by a combination of liquidity premiums, scarcity expectations, and state-owned backing.

Multiple sources indicate that StepFun plans to submit its listing application to the Hong Kong Stock Exchange in the first half of 2026. Choosing Hong Kong benefits from the institutional inclusiveness of Chapter 18C for specialized technology companies, facilitates the exit demands of domestic RMB funds, and avoids geopolitical uncertainties associated with U.S. listings. Currently, the company has engaged intermediaries to advance structural reorganization and financial sorting, but as of May 20, 2026, no formal A1 application documents have appeared on the HKEX's Disclosure Easy platform. Compared to the already-listed Zhipu (listed on January 8, 2026) and MiniMax (listed on January 9, 2026), StepFun's filing pace is more compact (compact/intense). If Hong Kong liquidity fluctuates in the second half of the year or regulatory inquiries delve deeply into data compliance, related-party transactions, and customer concentration details, technical delays in the listing process remain possible. Capital market patience is engaged in a two-way game theory (game/interaction) with the company's preparation progress.

The Deep Zone of 'AI + devices': Hidden Concerns and Breakthroughs Amid Commercialization Exuberance

StepFun's IPO preparation coincides with a profound reconstruction of narrative logic in China's large model industry. Over the past three years, capital chased 'who could build China's OpenAI,' but some open-source models achieved performance leaps at relatively low training costs, superposition (combined with) internet giants' overwhelming traffic and ecosystem dominance in C-end applications, narrowing the entrepreneurial window for pure general-purpose large models. The industry landscape has clearly stratified: the top tier is controlled by open-source foundations and superplayers, while the middle tier differentiates into enterprise services, vertical industries, and device entry points. StepFun has chosen to bet on 'AI + devices,' attempting to become the 'underlying brain' for smart hardware. This path effectively avoids the cutthroat competition in the C-end red ocean but ties its fate to hardware shipment cycles and the depth of system-level cooperation with manufacturers.

While the 'AI + devices' narrative offers industrial depth, commercialization still faces practical challenges. First, revenue structure may rely heavily on a few leading smartphone and automotive manufacturers. Periodic fluctuations in the hardware market, manufacturers' tendencies toward in-house model development, and prolonged terminal replacement cycles may squeeze third-party AI suppliers' bargaining power and revenue-sharing ratios (industry norms often range from 30%-60% based on token volume or revenue sharing). Second, the disruptive risks of technological routes have not dissipated. Large model competition is shifting from 'parameter scale races' to 'lightweight on-device deployment and scenario adaptation.' If StepFun fails to continuously iterate in vertical industry know-how accumulation and low-power inference optimization, its technological moat may be quickly eroded.

Moreover, the Hong Kong market's valuation logic for AI companies is trending toward rationality, with investors focusing more on operating cash flow and paid conversion rates. If StepFun relies solely on revenue growth to justify its high valuation, post-listing Valuation Regression (valuation reversion) to fundamentals seems inevitable. Caught between dimensionality reduction by giants and peer differentiation, StepFun's listing bell may merely mark the starting line of this prolonged commercialization marathon. What's your take? Discuss in the comments. (Produced by Thinker Finance)

Source: Investor China