Yin Qi, the veteran founder of Megvii whose company failed to achieve the long‑awaited IPOs, has quietly reassembled his ambitions into a far larger enterprise: a linked cluster of large‑model AI development, smart driving systems and an auto‑hardware ecosystem anchored by Geely. In less than three years the start‑up StepFun (Jieyue Xingchen/阶跃星辰) has gone from a low‑profile model developer to the recipient of a blockbuster B+ funding round—over RMB5 billion—backed by state investors, Tencent and existing VC backers. Yin’s formal installation as StepFun chairman on the day of the fundraising signalled both his personal pull with capital and a deliberate pivot by investors toward projects with clear industrial pathways.
That pathway is explicit. Yin also controls Qianli Technology (formerly Lifan, now in the Geely orbit), a company positioning itself as an auto supplier of smart driving and cockpit systems. StepFun’s software is already embedded in Geely’s Galaxy M9 through an AgentOS smart‑cockpit deployment, and the groups have announced a staged roadmap from L2+ systems to an L3 autonomous offering next year. By knitting together model development, vehicle subsystems and an OEM channel, Yin hopes to realize the ‘‘closed‑loop’’ commercialisation that many standalone model firms have so far struggled to deliver.
The timing explains why investors were willing to plough capital into StepFun even as overall funding for model vendors dried up last year. In 2025 the so‑called model layer saw just 22 deals worth about RMB9.4 billion, roughly half the prior year’s activity, reflecting a tougher, more discerning funding environment. After the early era when a coat of ‘‘AI’’ could unlock easy cheques, money is now flowing to companies that can couple models with real revenue, differentiated product features and defensible enterprise relationships.
Yin’s personal narrative matters to that calculation. A decade and a half of building vision‑systems and fighting for commercial traction have left him with credibility and scars. Megvii—once among China’s AI vanguard alongside SenseTime and Yitu—saw H‑share and A‑share listing plans collapse, and its founders paid the price in deferred exits and layoffs. Yin has repeatedly framed his new strategy as learning from those failures: AI1.0 was ‘‘hammer looking for nails’’; AI2.0 must show closed loops and hard monetisation.
The Geely tie‑up gives Yin a powerful springboard but also a looming constraint. Geely’s sales scale—several million cars a year across brands—means immediate routes to deployment and recurring revenues, and a chance to amortise software development across large production volumes. Yet industry insiders note the strategic dilemma: how many rival OEMs will hand over their vehicles’ ‘‘brain and eyes’’ to a supplier so clearly dominated by one carmaker? The Geely halo may therefore double as a cage, reducing StepFun’s bargaining power with potential external partners and capping its addressable market as a purportedly independent model vendor.
StepFun’s commercial choices have already shifted under Yin’s influence. The company abandoned a consumer‑facing play—social apps and AI chat products—after poor traction, and refocused on embedded multimodal models and vehicle solutions. The firm markets itself on fast base‑model iteration and multimodal capabilities, a technical identity that now combines with tangible in‑car billing lines: StepFun estimates modest but meaningful per‑car product revenue from cockpit OS deployments in Geely models. For investors chasing predictable revenue growth, that narrative is persuasive.
But it is far from guaranteed. A standalone independent model market still exists for a handful of players—ByteDance and Alibaba among the big tech incumbents, and firms such as Zhipu, MiniMax and DeepSeek for the independent camp—creating a tight competitive set sometimes referred to as a ‘‘base‑model five.’’ The survivors will likely be those that can either leverage proprietary data and distribution or embed into hardware ecosystems with recurring monetisation. Yin’s model chooses the latter, betting that hard assets and OEM relationships will outlast the froth in pure software plays.
If StepFun succeeds, the prize is significant: an AI software vendor with direct lines into auto production, scaleable unit economics, and a plausible pathway to public markets. StepFun already plans a pre‑IPO round and aims for a listing this year, which would make it among China’s first public pure‑play model vendors. If it fails to demonstrate independence or profitable growth, however, its market valuation and long‑term trajectory will be limited by Geely’s ownership patterns and the same commercial realism that humbled many AI pioneers in the last cycle.
Yin’s recent partial sell‑down of Qianli shares to raise cash and invite strategic partners signals awareness of this tension. The move frees funds and creates space to bind new industrial investors to the project, but it also underscores that the enterprise is still being engineered: capital, partners and corporate structures must line up before the ‘‘closed loop’’ can convert into sustainable margins. For global observers, Yin’s wager is a revealing test of the Chinese AI playbook: whether national champions are built by platform‑ecosystem giants or by coalitions that combine models, devices and domestic industrial partners.