Horizon Robotics: Driving the Autonomous Revolution
Scottish Mortgage is Baillie Gifford's flagship investment trust (founded 1909, LSE ticker SMT), known for its maximalist growth style — long-term stakes in Tesla, Amazon and ASML plus bold allocations to private companies like SpaceX and ByteDance. It is the UK retail investor's flagship vehicle for global disruptive growth.
In plain words
This interview covers Horizon Robotics, a Chinese company that designs self-driving chips and software. Its founder predicts that within three years you won't need to hold the wheel on daily commutes, within five you won't need to watch the road, and within ten you can sleep in a moving car. He argues hardware and software must be developed together, and China will be the first big market. For everyday investors, it's worth reading because it explains why Volkswagen put $2 billion into the company—and flags real risks: carmakers might try to build their own systems, and global expansion is uncertain.
This report covers a Scottish Mortgage podcast interview in which Dr. Kai Yu, founder and CEO of Horizon Robotics, argues that cars are transforming from mechanical machines into "computers on wheels." The company is dedicated to autonomous driving solutions that integrate software and hardware, and
At a Glance
Guest: Dr. Kai Yu (余凯), founder and CEO of Horizon Robotics, a 30-year veteran in machine learning, and founding director of Baidu’s Institute of Deep Learning (IDL). The conversation was hosted by Lawrence Burns, portfolio manager at Scottish Mortgage. Its focus was to dissect Horizon’s technology roadmap, market position, and investment logic.
Main thesis: Starting from the premise that “a software company or a hardware company alone cannot solve autonomous driving,” Yu explains why Horizon chose an integrated software-hardware approach. He details the uniqueness of the Chinese market, the “Journey Together” partnership philosophy, and the platform ambition of extending from vehicles to general-purpose robots. Burns, from an investor’s perspective, supplements with the rationale for holding the stock and the risks involved.
Most consequential statement in the episode: Yu provides a clear timeline — within 3 years, 100% hands-off for daily commutes; within 5 years, 100% eyes-off from the road; within 10 years, fully “mind-off” driving, by which point a person can sleep inside a moving car. At the same time, he calls himself a “conservative optimist” and explicitly believes RoboTaxi will not be deployed at scale in the near term.
Key Themes
1. Software and Hardware Must Be Co-Designed — After Baidu’s Rejection, He Chose to Go It Alone
The core of Yu’s argument: To build dedicated hardware for autonomous driving, one must first deeply understand autonomous-driving algorithms; building hardware first and then waiting for software companies to adapt is too slow — software and hardware must be designed together. He uses this logic to explain Horizon’s differentiated footing among the many players — “Among all hardware companies, we are the strongest in software; among all software companies, we are the strongest in hardware.”
This assessment stems from three experiences. First, in 2010 he led a team to win first place in the first ImageNet challenge; by 2012, when Geoffrey Hinton won the third ImageNet, participants generally relied on GPU computing power. At the time, Yu’s Baidu laboratory was already one of NVIDIA’s largest customers for deep-learning GPUs, and he judges that Jensen Huang himself had not yet realized the potential of GPUs for machine learning, only truly waking up around 2014. Second, he launched China’s first autonomous-driving project within Baidu, but the test vehicle’s trunk was stuffed with cables and equipment, and after 30 minutes it overheated and had to stop to cool down — “extremely unreliable.” Third, he proposed to Baidu’s senior management that the company develop dedicated hardware in-house, but Baidu’s culture was “extremely software-driven,” and the broader Chinese industry at the time believed that hardware development had too long a payback period; “no one wanted to touch hardware.” So he chose to leave and founded Horizon in 2015.
From this he reasons backward about the industry landscape: many semiconductor companies “went down the wrong path” — first making the chip and then waiting for software companies to adapt; traditional automotive chips often take 5 to 10 years from tape-out to installation in a vehicle, which is too slow. Horizon develops software and hardware in parallel, moving much faster than all competitors, and “in a very short time, we, as a new entrant, became the leader in China.”
Burns adds: This software-hardware co-design capability is Horizon’s core moat; its complexity means that in the long run it is very difficult for automakers or Tier 1 suppliers to fully replicate such capability in-house — this is simultaneously a source of investment logic and risk.
2. China Is the “Smartphone-Style” Starting Point for Autonomous Driving — From First Supplier to Global Diffusion
Both guests share the assessment: The crowding and competitive intensity of China’s EV market are unique globally, and precisely for that reason it has become the soil where intelligent driving technology is first commercialized at scale. Burns’s exact words: “No other market is like China” — extremely crowded and extremely competitive, yet “leading the global EV race with quality and incredibly low prices.” Yu adds three conditions that make China the launch market: users are open-minded toward new technology; the degree of digitalization in society is higher than in most Western countries; and EV and intelligent-cockpit penetration far exceeds other regions.
The data supports the “China first” position: Last year (2023), 3 million new cars sold in China came pre-installed with Horizon solutions; its current market share in China exceeds 50% (Burns’s figure); almost all major OEMs in China are customers. The globalization path relies on two channels: direct cooperation with foreign brands such as Volkswagen, and the overseas expansion wave of Chinese EV makers themselves. Volkswagen’s USD 2 billion investment is seen by Burns as a critical asset — both a certification of technical capability and a channel into global markets. Yu sums up the cooperation as “German quality with Chinese speed,” but he emphasizes that the root of trust is the “Journey Together” philosophy: the goal is long-term win-win; “it is not a party, but a long journey.” In an industry transition, many things cannot be delivered in a standardized way; “Horizon and the OEM must define them together.”
Yu’s analogy for the future diffusion path is the smartphone: first widely accepted in China, then gradually fully embraced in regions such as Europe and Japan. “Intelligent EVs” and “computers on wheels” will not belong to China alone — China is only where it first happens; once the technology matures, it will become the next global trend. He also lists the global Tier 1 partners: Bosch, Continental, Denso, ZF.
Readers should be reminded: Yu’s narrative that “the China experience is globally replicable” carries the coloring of a position-holder’s perspective. The interview did not discuss specific geopolitical constraints on the globalization path, with Burns merely noting that “almost all markets except the United States are reachable.”
3. A Ten-Year Blueprint: From “Hands-Off” to “Mind-Off,” Then to a Platform for All Robots
Yu’s view of the autonomous-driving evolution: Over the past four or five years, the Chinese industry has moved from limited scenarios such as highway driving, lane keeping, and AEB to full-scenario autonomous driving in busy urban traffic. End-to-end large-model training has made technical progress “accelerate significantly year after year,” and the inflection point is this year or next. His specific predictions:
| Timeline | Goal | Meaning |
|---|---|---|
| Within 3 years | 100% hands-off | Daily commutes with no need to hold the steering wheel |
| Within 5 years | 100% eyes-off | No need to watch the road during the entire drive |
| Within 10 years | 100% mind-off | Can sleep in the car; no mental engagement throughout |
He labels himself a “conservative optimist” and stresses that RoboTaxi will not arrive this year or next — because RoboTaxi is by definition mind-off and requires more time to resolve edge cases. And the path is “step by step, building a very solid business along the way.”
On the long-term vision, Yu argues that the car is merely the first application of robotics technology, and “the computing paradigm is almost the same” — environment perception, mapping and localization, human-machine interaction, decision-making in complex environments, and control; these capabilities can be transferred to scenarios such as elderly care, home services, agriculture, manufacturing, and logistics. At the same time, he acknowledges two uncertainties: humanoid robots have a higher degree of freedom and complexity than cars, requiring breakthroughs in software and theory; and energy consumption is a bottleneck — the human brain runs on just 20 watts, while current computing architectures consume far too much energy, possibly requiring a shift to “biologically inspired software and hardware architectures,” which will take 10 to 20 years of sustained progress. His ultimate statement: “I want to be like Microsoft for robots” — in other words, he wants to become the Microsoft of the robotics field.
Burns translates this vision into investment language: Horizon’s opportunity comes in two steps — the first is “turning cars into AI robots,” and the second is expanding its capabilities to a broader range of robot categories. “The biggest outliers are companies that use one thing to do another thing,” and Horizon has the potential for such a “second act.” He cites chip shipments as evidence of the room in the first step: last year Horizon shipped 2.9 million chips, and this year it plans to ship more than 10 million; China sells roughly 30 million cars a year, and the world outside the U.S. sells roughly 60 million, so the market is far from saturated; as autonomous-driving functions strengthen, the average selling price of chips could also rise.
4. An Investor’s Perspective: Eight Years of Relationship Building, and Two Main Risks
Burns reviews Scottish Mortgage’s position-building process: As early as 2015, the year Horizon was founded, he had visited the company on site with Linda Lin (head of Baillie Gifford’s China operations) and former portfolio manager James Anderson; at the time Horizon was still more focused on chips and robotics. They tracked it for years and finally invested in 2021. He describes Yu as “one of the world’s leading AI scientists,” noting that technical capability itself attracts top talent, and that he is “not someone who jumped onto the AI wave only in the past three or four years” — this is a lifetime’s work.
Burns identifies two main risks on his own initiative: First, large automakers may attempt in-house development to save costs; but “mastering both chips and algorithms and achieving vertical integration is extremely difficult,” and “at the very least, not every brand is suited to in-house development” — he judges the more reasonable path is partner enablement, and Scottish Mortgage hopes that partner is Horizon. Second, globalization challenges — outside the United States, Horizon has the opportunity to reach most markets; the Volkswagen investment is a “particularly important channel,” and Yu himself has already thought deeply about the company’s identity and global footprint.
It should be noted here that Burns, as a portfolio manager with a stake in the company, has a judgment — “in-house development is extremely difficult, partner enablement is more reasonable” — that is aligned with his investment position; readers should regard it as the view of a position-holder.
Companies Mentioned
| Company | Guest Stance | Key Data |
|---|---|---|
| Volkswagen | Positive / Core partner | Invested USD 2 billion; “German quality with Chinese speed” |
| NVIDIA | Neutral / Dual role as competitor and supplier | Yu’s team was once one of the largest customers of deep-learning GPUs; Jensen Huang only realized the machine-learning potential of GPUs around 2014 |
| Huawei | Risk flag / Competitor | One of China’s largest private enterprises; competes with Horizon |
| Baidu | Neutral / Former employer | Yu founded China’s first deep-learning laboratory (Baidu IDL, 2013) and China’s first autonomous-driving project (2012) |
| Bosch, Continental, Denso, ZF | Positive / Global Tier 1 partners | No specific data provided |
| Microsoft | Neutral / Analogy | Yu’s goal: “Microsoft for robots” |
| Siemens, NEC Labs | Neutral / Career history | Yu’s first and second jobs |
| Alibaba, Tencent | Neutral / Contemporary context | Mentioned as Chinese internet giants that also lacked machine-learning teams at the time |
Judgments Worth Remembering
1. Kai Yu: “I label myself always as a conservative optimist.” — He offers a 3/5/10-year three-tier autonomous-driving timeline, but explicitly believes RoboTaxi cannot be solved in the near term.
2. Lawrence Burns: Horizon’s opportunity is “the opportunity to both create the hardware and the operating system of AI” — the first step is making the car an AI robot, and the second act is extending to general-purpose robots; “the biggest outliers are companies that use one thing to do another thing.”