NeuroAI NEUROAINEUROAI.SITE
ESC

Yu Kai's Horizon Robotics: from Baidu lab to a HK$5.4 billion Hong Kong debut

How machine-learning scientist Yu Kai left Baidu, founded Horizon Robotics in 2015, and took China's largest smart-driving chip maker public in Hong Kong.

2026-10-25 · 811 words · NeuroAI
Yu Kai's Horizon Robotics: from Baidu lab to a HK$5.4 billion Hong Kong debut

A self-driving car needs to perceive the world in milliseconds, but the chip that does the seeing is easy to overlook. The company that supplies a large share of those chips in China was started by a researcher who had never built hardware.

Yu Kai (余凯) spent two decades in software before deciding the bottleneck was silicon.

Key takeaways

  • Yu Kai (余凯), born in 1976 in Nanchang, Jiangxi, founded Horizon Robotics (地平线) in July 2015 after leading Baidu's Institute of Deep Learning.
  • He holds a PhD in computer science from the University of Munich (2004) and degrees from Nanjing University; he previously worked at Siemens, NEC Labs America, and Stanford.
  • Horizon listed on the Hong Kong Stock Exchange on 24 October 2024 (code 9660.HK) at HK$3.99 per share, raising about HK$5.4 billion (≈ US$695 million) — among the year's largest tech IPOs there.
  • Its Journey (征程) chip family powers advanced driver-assistance systems; cumulative shipments passed 6 million sets by late 2024, used in roughly 290 vehicle models from 27 OEMs.
  • In China's 2024 market for smart-driving computing solutions for domestic-brand passenger cars, Horizon held the top share at about 34%.

A scientist who turned to silicon

Yu's career began far from the automobile. He earned bachelor's and master's degrees in electronic engineering from Nanjing University in 1998 and 2000, then a doctorate in computer science from the University of Munich in 2004. He worked as a senior researcher at Siemens and later led a media-analytics group at NEC Labs America, and taught as an adjunct at Stanford. His research centered on deep learning years before it became mainstream.

In 2012 he returned to China to join Baidu, where he founded and led the Institute of Deep Learning (深度学习研究院, IDL) and launched one of the country's earliest autonomous-driving projects in 2013. By 2015, watching algorithms run dozens of times faster on specialized hardware, he reached a conclusion: the next leap needed a chip built for neural networks, not a general-purpose one.

Betting on the hard problem

In July 2015, at 39, Yu left Baidu and founded Horizon Robotics in Beijing with two former colleagues. The thesis was unfashionable: build a car-grade ("车规级") AI chip when most Chinese chip effort went into phones or cloud. He later framed the strategy as competing "where no one else competes" — picking the hardest, least crowded lane.

The payoff took years. Horizon released its first Journey processor in 2017 and, in 2019, China's first mass-produced automotive-grade smart-chip, the Journey 2. From there it scaled into a full stack of software and hardware for advanced driver-assistance (ADAS) and higher-level autonomous driving.

The public debut

On 24 October 2024, Horizon listed on the Hong Kong Stock Exchange's main board under code 9660.HK. Priced at HK$3.99 per share, the offering raised about HK$5.4 billion (≈ US$695 million) and was described by multiple outlets as the largest technology IPO in Hong Kong that year. For a company that started as a research bet, the listing was a signal that automotive AI chips had become a mainstream, fundable category.

Scale on the road

Horizon's numbers show how fast the category grew. By August 2024 its Journey chips had shipped more than 6 million sets, equipping about 290 models from 27 OEMs; all of China's top ten OEMs had adopted its solutions. In 2024 it led China's market for smart-driving computing used by domestic-brand passenger cars with roughly a 34% share. The chips sit inside the assistance systems of mass-market vehicles, not just premium ones — the quiet infrastructure behind features drivers now take for granted.

Why it matters

Yu's story is about the migration of AI from cloud to edge. Rather than training models in data centers, Horizon puts inference on a chip inside the car, where latency and safety leave no room for round-trips to the cloud. That edge-AI approach is now spreading from vehicles to robots, a direction Horizon has publicly begun to pursue.

Honest limitations

Figures on shipments, OEM count and market share come from company statements and state-media business reporting around the IPO; they reflect a point in time and will have moved since. Birth year and early biography draw on Chinese profiles and news features that broadly agree but are not primary documents. I have not independently verified every financial metric and have omitted later quarterly results. The "largest tech IPO" characterization refers to Hong Kong listings in 2024 per reporting at the time.

What readers can do now

  • If you buy or spec a car in China, check whether its driver-assistance system runs on Horizon Journey silicon — it likely affects update cadence and feature depth.
  • Investors can read Horizon's HKEX prospectus to separate recurring royalty revenue from one-off hardware sales.
  • Engineers interested in edge AI should study the Journey architecture as a case of co-designing chip and software for a single task.

Related coverage

Cite this article

NeuroAI (脑机网), "Yu Kai's Horizon Robotics: from Baidu lab to a HK$5.4 billion Hong Kong debut," 25 October 2026, https://neuroai.site/a/na-people-yu-kai-horizon-founder

Machine-readable: BibTeX · RIS · JSON · Markdown

More in “Builders” → · Back to home · Markdown version