Yin Qi: On Chairing StepFun, the Seduction of Intelligence, and a Brutal Long-Horizon Game

Ying Qi, a seasoned AI entrepreneur with nearly 15 years in the trenches, recently stepped into a new role. Already chairman of Qianli Technology, an A-share listed company focusing on AI and cars, he is now also taking the helm of Jiyue Xincheng, a foundational model startup. In a candid interview with Xiaojun, Ying reflected on the fierce transition from AI 1.0 to 2.0 and the strategic bets behind his expanding portfolio.

He opened with a stark observation: “This is probably the most intense tech competition in history.” Having matured from the “play every hand” mentality of youth, he now knows that some cards aren’t worth playing. An IPO, for instance, is merely a waypoint for survival, not a magic fix. The real challenge is building a sustainable business in an arena where foundation models demand insane capital—at least 3 billion yuan per year, potentially over 10 billion in three to five years. With that kind of burn, Ying argues, pure B2B services or B2C software apps are dead ends for a startup. B2B yields thin margins and long sales cycles; B2C pits you against internet giants with built-in data flywheels. His solution: tightly couple AI with hardware terminals. By embedding models into devices like autonomous vehicles, robots, or yet-to-be-invented gadgets, startups can compete in a market that isn’t winner-take-all, and where hardware-software-AI integration creates a defensible moat.

This conviction is rooted in his journey. Ying co-founded Kuangshi Technology, a pioneer in computer vision, and watched the industry shift overnight with ChatGPT. Initially, he dismissed large language models as a niche evolution, but after deep tinkering and internal debates, he recognized it as a tectonic shift—an “unavoidable battle.” Thus Jiyue was born in April 2023, its English name Step Function symbolizing the leap from quantitative change to qualitative transformation, much like the nonlinearities that define deep learning. From day one, the mission was twofold: be among China’s top foundation model companies, and achieve commercial closure through AI-terminal synergy.

The road is brutal. Ying doesn’t sugarcoat: talent wars, capital intensity, and relentless technological churn make this the “fastest, biggest” race he’s seen, complete with bubbles in stock prices and developer salaries. The elimination round is past halftime; fewer players will survive. Yet Jiyue intends to stay the course—pushing the frontiers of foundation models for at least three more years, while letting real-world terminals “pull” its R&D. That means prioritizing practical metrics like efficiency and cost over benchmark bragging rights.

Wrapping up, Ying positions himself as an “AI-native entrepreneur,” allocating 80% of his time to tech and product, regardless of titles. Qianli handles the “body”—cars and mobility platforms—while Jiyue powers the “brain.” This brain-body fusion aligns with his long-held belief, inspired by Jeff Hawkins’ On Intelligence, that AGI must interact with the physical world. You can’t skip spatial survival and jump straight to high-level reasoning. For Ying, the hard path through hardware is not just a business model; it’s the only way to build intelligence that truly matters.