Corporate Q&A

In the driving seat

WiC’s exclusive interview with the founder and CEO of Horizon Robotics


Last week Week in China published the third update of our Top 50 China Unicorns ranking. Appearing again – with a valuation of $5 billion – was Horizon Robotics. Only six years-old, the Beijing-based firm is a leader in AI (artificial intelligence) chips for applications in autonomous driving.

So how will self-driving cars transform the future? We spoke to Yu Kai, the founder of Horizon Robotics, who left Baidu in 2015 to set up his pioneering venture in China after a career in Silicon Valley.

Can you talk about the benefits the world will derive from driverless technology?

Our mission is really centred on the idea of making mobility safer for everyone and liberating drivers from the stressful parts of the driving experience. Do you know that every year 200,000 people die in car accidents in China alone? I think we are solving a very important problem in a very fundamental way, because our systems are designed not only to make a journey more efficient, but also to save lives by significantly reducing accidents. Efficient mobility can also cut down congestion and energy consumption.

Of course, our business is all about creating positive interactions between human and cars. We don’t intend to take the driving out of the equation like RoboTaxis.

When do you predict driverless cars will become the main transportation in Chinese cities? Put another way, if you had a six year-old child today, would you expect they will need driving lessons?

The development of autonomous driving is not easy, but its market potential is huge. Last year China started mass producing assisted-driving vehicles and we all saw a significant growth in this market. I predict that vehicles equipped with Level 3 systems will take up a big share of the auto market by 2025. Truly driverless vehicles will first be introduced to business settings, such as in ports, docks and mines, also starting in 2025. But a truly consumer-facing rollout will take much longer time. [Editor’s Note: Level 3 status is defined by JDPower as “conditional driving automation”: “People inside the vehicle do not need to supervise the technology, which means they can engage in other activities. However, a human driver must be present, alert, and able to take control of the vehicle at any time, especially in the case of an emergency due to systems failure”.]

You have accomplished a lot in a short period of time. How do you describe your company today and what’s your mission?

I’m very proud to say that Horizon Robotics has become not only the first Chinese company to commercialise autonomous driving processors, but also one of only three companies in the world to have production-grade ADAS [Advanced Driver Assistance System] processors. We are also the ONLY company that offers a full set of AI solutions facilitating Level 2 [where drivers need hands on the steering wheel but the AI actively steers, accelerates and brakes the vehicle, to quote JDPower’s explanation again] through to Level 4 autonomous driving right now [L4 requires no human interaction as the vehicle is programmed to stop itself in the event of system failure]. We are well-known in the industry for providing top class Edge AI computing platforms for intelligent vehicles. In terms of our mission, we want to make human life safer and better through Edge AI computing, and our vision is to empower every vehicle with our Edge AI computing platform.

What is Edge AI and why is it so important for self-driving cars?

Edge AI means that all the AI computing performance required to realise a function needs to be available readily from the edge device locally.

This is not only a market trend but also our vision. A few years ago, cloud was the trend because people liked the idea of uploading all the data and conducting computing in the cloud. However, today people value timely response, reliability and privacy more than convenience. This is especially true in autonomous driving. Take road safety as an example: if a child is crossing the road, you wouldn’t want to rely on cloud computing for driving instructions because there could be network problems that delay data transmission. With local computing, the data processing and instructions are more instantaneous and reliable.

The second advantage of edge computing is privacy protection. If voice instruction devices use cloud computing, the data needs to be uploaded to the cloud for identification and that process risks data leaks.

We believe Edge AI will continue to grow, as processors like our Journey Series become more powerful, energy-efficient and consume less computing power, thus lowering the cost.

In making chips for Edge AI applications, would you describe Horizon Robotics’ core expertise as hardware design or software design?

When targeting domain specific AI processing, as we do for automotive intelligence applications, it is essential to design the architecture of the processor with application software requirements that maximise performance while reducing cost. Therefore, we don’t see processors and algorithms as separate fields and our strength is exactly the ability to leverage our expertise in both in seamlessly integrating hardware and software architectures.

That is how we were able to design the most efficient Edge AI automotive processors for our clients.

One of our key offerings today is the open platform model which provides the baseline algorithms in the AI processors to all customers. Then we tweak the rest of the software to meet their needs. In this sense the software and hardware work together to yield the smoothest and safest operation.

Our approach to OEM partners is to understand their needs first, then work with them to pick and choose product features. After that we design the AI solutions accordingly. We provide both software and hardware, but we give our clients the option to choose whatever they want and however they want it.

Who are your main customers? How do you work with them?

As a tier-two auto supplier, Horizon Robotic’s customers include both tier-one auto suppliers and car companies or OEMs. We are fortunate that most Chinese automakers have the ambition to go “intelligent”. The trend is unstoppable! However, the intelligent automotive processor is hard to develop. That puts us in a sweet spot as we can provide AI solutions that are super-efficient, user-friendly, safe and reliable, and cost-effective.

We currently have partnerships with about 80% of the major Chinese OEMs, including BYD, Chang’An, Great Wall Motor, Li Auto, SAIC and Guangzhou Auto. We provide AI computing platforms with reference models, toolchains and algorithms on which they can build to cater to their own needs. We empower them from the bottom up and have achieved long-term collaborations.

The automotive industry supply chain is undergoing disruptive changes. The old pyramid-shaped hierarchical structure is now replaced by jigsaw-style partnerships because of the importance of speed. Today a company is no longer allowed to innovate alone at its own pace. Collaboration is essential!

Who are your business partners? Any well-known international players?

Our ecosystem is global, wide and expanding. We are actively connecting with key players in the automotive industry. We recently announced our joint venture with German auto tech company Continental. This partnership will see our AI processors designed into Continental’s smart cameras and scalable Automated Driving High-Performance Computer series. What’s exciting about this Shanghai-based JV is that it will supply the finished products not only to Chinese OEMs but also to brands across the world, giving us a global reach and manufacturing footprint.

You left a senior role at Baidu to start Horizon Robotics in 2015. Can you tell me a bit about your background and what inspired you to make the leap?

I had worked at Baidu for over three years but my passion has always been machine learning since my school years. I even taught ‘Introduction to AI’ at Stanford University and University of California – Santa Cruz before joining the corporate world. While working at Baidu in Silicon Valley and later in Beijing, I also led the Institute of Deep Learning and later its autonomous driving and image research division.

I still remember when I first saw one of the earliest autonomous driving prototypes. I was shocked to see huge machines connected by messy cables in the car’s trunk. During the test drive, the machine got overheated so easily that it had to stop every half hour to cool down. That made me think: if only I could design a computing hardware that was efficient, compact and low on power consumption, then I could probably solve a fundamental problem. So, you can say that the desire to provide full solutions to future intelligent vehicles inspired me to set up Horizon Robotics. I was lucky that my two co-founders also worked at Baidu back then and shared my passion.

Where are you based now and how big is your team?

I’m in Beijing. Horizon Robotics has over 1,000 employees across the country, mostly in Beijing, Shanghai, Shenzhen, Nanjing and Hangzhou. Over 70% of them work in engineering and R&D.

Horizon Robotics completed a $900 million Series C financing earlier this year at a $5 billion valuation. And you are one of our Top 50 China Unicorns. What did investors say they liked most about the company’s prospects?

I cannot confirm your valuation number but I thank you for including Horizon Robotics in your Top 50 Unicorn list. I believe developing AI processors really needs long-term commitment. Over the years, I’ve found that the more we focus on platform empowerment and standardisation, the bigger the growth we achieve. And while we don’t make integrated circuits nor chase “quick money”, we gain more friends and business partners. Our investors trust us. I sometimes joke that they are like Zen masters who are always calm and long-term focused.

I am very grateful to them for sharing our long-term commitment. There are already many companies that are good at making quick money; we are among the few who don’t mind making slow money.

Horizon Robotics is a market leader in AI chip design. Can you give a quick outline of what makes AI chips different to other types of integrated circuit design?

AI chips are specialised silicon chips, which incorporate AI technology for machine learning. They’re more efficient than general purpose chips in solving mathematical and computational problems. By integrating hardware and software, we have developed a ‘Brain Processing Unit’ or BPU, which is optimised according to the needs of the applications. This is especially suited for autonomous driving scenes and functions.

People tend to think the bottleneck of an autonomous driving breakthrough is lack of computing power. Actually, computing efficiency matters more than computing power. For example, a car’s acceleration speed is a better indicator of its real power than its horsepower. Similarly, accurate FPS or “frames per second” is a better indicator of the AI chip’s performance than computing power, which is measured in TOPS or “tera operations per second”. I’m happy to say that our chips have the ultimate performance efficiency.

Horizon Robotics mostly makes ASIC chips. What are the main benefits of these kinds of custom chips compared to other formats?

ASIC chips are custom-made and designed for a particular application. In our case, we design chips almost exclusively for intelligent vehicles. In July this year we launched our third-generation AI processor Journey 5, which was designed for a high level of autonomy that delivers 128 TOPS processing capabilities for a single processor and consumes less power.

With Journey 5, we can not only introduce a range of intelligent driving central computers with 200 to 1000 TOPS AI computing power, but also achieve the industry’s highest FPS performance with very low power consumption.

We move one step closer towards full autonomous driving. Journey 5 is the first central computing AI processor that integrates autonomous driving and the intelligent cockpit. Its single chip has AI computing power that supports more than 16 high-resolution cameras, and meets the production demands of high-level autonomous driving.

There are a lot of players claiming track records or announcing plans to be a major force in autonomous driving technology. Who do you see as your main competitors, both in China and internationally?

Autonomous driving is the future. This field has become more exciting and dynamic with constant new entrants in China and internationally. However, this is still a new field, and we are all students trying to learn and perfect the technology. Horizon Robotic’s strength is the ability to provide a high-performance, high-computing platform that meets the needs of all levels of automated driving, a full range of solutions covering L2-L4 levels of autonomous driving, an open and easy-to-use processor and toolchains that fully enable our customers to create, upgrade and evolve efficiently.

How do you see the landscape playing out? Will the ‘network effect’ mean only a couple of players will thrive and the bulk will disappear?

Competition in the chip industry is fierce and is subject to the winner-takes-all philosophy. If you can’t be in the industry’s top two, you basically have no future. Even the number one and number two spots can face huge distances between them, just like the evolution of the PC and smartphone industries in recent decades.

I think the next three years is the key. We hope to get into the top three globally, or even the top two.

We do have first-mover advantage. The typical development cycle of auto-grade chips is five years and the later you get into the process, the higher the entry barriers you will face.

Having said that, we are also aware that we are only three years ahead of the game so we mustn’t be complacent. We need to continue making breakthroughs in order to maintain our leading position.

Click here to visit our dedicated Top 50 China Unicorns site for more background and insights on the companies contained on the list, including Horizon Robotics.

© ChinTell Ltd. All rights reserved.

Sponsored by HSBC.

The Week in China website and the weekly magazine publications are owned and maintained by ChinTell Limited, Hong Kong. Neither HSBC nor any member of the HSBC group of companies ("HSBC") endorses the contents and/or is involved in selecting, creating or editing the contents of the Week in China website or the Week in China magazine. The views expressed in these publications are solely the views of ChinTell Limited and do not necessarily reflect the views or investment ideas of HSBC. No responsibility will therefore be assumed by HSBC for the contents of these publications or for the errors or omissions therein.