Xu Bing, co-founder and executive director of SenseTime, attended the Hong Kong FinTech Week event and had a live conversation with former Forbes reporter Olivia Kinghorst. During the nearly 20-minute conversation, the two parties discussed AI cloud platform market trends, investment judgments, super applications, competitive landscape and other topics.
Xu Bing believes that humankind's demand for AI and computing power is rising rapidly. Three major fields, including personalized AI assistants, autonomous driving and robots, are expected to produce super applications and bring huge opportunities to the industry. The growth in demand for AI computing power is inevitable, and the key is to reduce costs. Based on the trinity strategy of “computing infrastructure – large models – applications”, SenseTime is building one of China’s largest AI cloud services.
In his view, not investing in infrastructure construction such as AI cloud platforms may mean missing important opportunities. This risk is far greater than the initial investment cost.
The existing trillion-dollar data center market needs to be redefined and restructured. SenseTime is currently the only large-scale AI cloud platform on the market that is independent of Internet giants, and has full-stack capabilities from computing infrastructure, large models, to applications, providing enterprises and research institutions with a solid and secure AI cloud. Service support.
Xu Bing said that we may be at the best starting point in history and in what may be the largest wave of technological change in history. This means that the more you try in the right direction, the more likely you are to succeed. SenseTime needs to be more adaptable, more flexible, and able to transform quickly.
The following is the transcript of Xu Bing’s interview at the event:
Moderator: Can you talk about what you and SenseTime are doing recently? Where do you think the biggest new opportunities in artificial intelligence currently lie?
Xu Bing: We are building one of the largest AI cloud platforms in China. Our goal is to achieve seamless integration of large-scale computing devices, basic models, and AI applications. Our development history in the past ten years has been extraordinary - from a deep learning research laboratory to a leading company in computer vision, and now one of China's leading AI cloud service providers.
We have been working in the field of AI for ten years. Our main task now is to build an AI cloud and be fully prepared to ensure that the company can expand to general artificial intelligence (scale AI to agi). In the past decade, the computing resources required to produce intelligence have increased a million times. We started investing in building our own computing resources very early. In 2019, our first kilo-card GPU cluster was put into use. Now our largest AI cloud has access to more than 50,000 GPUs. We have become the AI cloud platform of choice for thousands of enterprises, startups, and research labs to train and infer AI models.
We see five huge market opportunities in the future:
The first is exponential growth: AI advances at an incredible pace, with computing power requirements growing tenfold every two years. These rapid improvements in software, hardware, and large-scale infrastructure are adding up, resulting in rapid improvements in training, post-training, and inference capabilities.
Second is the rapidly changing landscape: AI has moved beyond simple pre-trained models. We see developments in post-training, large-scale inference, multi-modal capabilities, synthetic data, and higher-order inference. As long as we are not limited by computing resources, there are many opportunities for innovation.
The third is the rise of inference: Inference is becoming increasingly important and complex, requiring a new cloud infrastructure deployed everywhere, from data centers to edge devices.
Video and multimodal models: The future of AI goes far beyond text models. We are working on video models and multi-modal models, which require more computational requirements.
The last and most important thing is killer applications like AI assistants: in the future we will see the booming development of AI killer applications. Imagine having a personalized AI that truly understands your habits, preferences and goals to help you navigate work and life. They may not be perfect at first, but they will improve over time. Imagine robots, living and working alongside us on this planet. These innovative applications are being developed on our AI cloud.
Moderator: In the capital market we have been debating an issue: whether building AI infrastructure and AI models can achieve sufficient ROI. Big tech companies and governments will spend approximately $1 trillion in capital expenditures on AI over the next few years. Will such investment costs be repaid?
Xu Bing: We need to realize that humankind’s demand for computing power and AI is constantly growing. Whether it’s the AI craze we’re seeing now, gaming, or just watching short videos. Everything requires more computing power and AI technology. Just imagine: if humans had one million times more computing power now, we might be able to compress 100 years of technological progress into less than 10 years.
Here comes a problem of calculating costs. Therefore, companies like SenseTime are needed to significantly reduce computing costs and make AI cheap and accessible. The current state of things is that AI is getting better and better every year, and the cost is actually getting cheaper and cheaper.
Since AI is the direction favored by everyone, powerful companies will definitely invest in advance; short-term costs are not that important. Because not investing could mean missing out on significant opportunities. This risk is far greater than the initial investment cost.
So to answer your question, first of all I see an existing trillion-dollar data center market being redefined and rebuilt. CPU-based data centers no longer generate attractive returns. However, as the demand for AI computing continues to explode, AI data centers (intelligent computing centers) have become a triple-digit growth market.
Secondly, I think the value of AI is extremely profound. Before reaching AGI, humans were already making good use of current AI to create value. You can talk to biologists, materials scientists, physicists, game designers, manufacturing engineers, and robotics engineers and ask them whether AI is already changing the way they work.
The point I'm trying to make is that we don't need to wait for some perfect, human-level AGI to emerge to see huge value returns. AI has improved the way people work and created value in countless areas.
Moderator: Now, let’s discuss your competitive advantages. You are a pioneer in building AI infrastructure in China and the first company in China to release a large GPT-4 level model. However, there are now many Internet giants and startups competing in this field. What do you think of your competitive barriers?
Xu Bing: In my opinion, SenseTime is unique. First, let’s take a look at China’s artificial intelligence competitive landscape. It is actually very similar to the West. The four major Internet giants (ByteDance, Alibaba, Tencent and Baidu) are all actively investing in AI and they have their own AI cloud infrastructure. With hundreds of thousands of GPUs, they try to develop their own chips, and they use more computing power and data to train models than their opponents. They own more than half of the computing resources on the market. Apart from these four giants, for many other players, computing resources are the bottleneck that limits the speed of innovation.
As for SenseTime, we are the only large-scale AI cloud platform independent from Internet giants. Our strengths go beyond training the best AI models. We have spent ten years building a complete AI technology stack – from hardware infrastructure, to AI cloud, to models and applications. This is an insurmountable barrier for most players. We focus on accelerating the entire AI development cycle, from data cleaning and preparation to training and inference. Every step is important and will be optimized over and over again. Moreover, our AI cloud is adapted to various GPU chips and can match the most suitable computing chips for different AI tasks.
Cloud infrastructure will fundamentally change as we live in a world where computing architecture is redesigned to serve the AI wave. The way we process information will fundamentally change in the future. In the past, cloud computing was designed based on retrieving pre-stored information—we wrote text, made videos, stored them, and then retrieved them using algorithms. In the future, information is generated by AI based on prompts. This is what we call generative AI. The future of computing will rely more on generation than retrieval. This is why the cloud of generative AI is a market from 0 to 1. This is something completely new. Our goal is to build a leading AI cloud platform and make it ubiquitous, not only in data centers but also on terminal devices. We are not trying to take market share from anyone. We are helping create this new market.
Moderator: We know that generative AI may impact various industries. What do you and your customers think will be the killer AI applications?
Xu Bing: China’s AI ecosystem is booming. We see thousands of AI startups emerging every year. These entrepreneurs come from all walks of life, including Internet companies, manufacturing, the electric vehicle industry, etc. They worked with AI scientists to form an amazing entrepreneurial team. You know how "volatile" Chinese entrepreneurs are, right? They are bringing this hard work to developing AI products, such as apps and robots. SenseTime has also invested in dozens of promising startups. We don't just invest cash, we also provide them with computing power and model capability support through our AI cloud. All this is to jointly promote the development of AI applications.
We have identified several super apps or killer apps that have the potential to create huge industry opportunities.
First there are robots. Today's robots are completely different from what we were talking about three or five years ago. The robotics industry has undergone a complete transformation in the past two years, and has even taken on a new name: "embodied intelligence." We need to consider the possibility that within our lifetimes humans will coexist with robots, and that their numbers may even exceed the global human population. I think the possibility is over 50%. We will share the planet with another intelligent species—one created by us, driven by AI, and manufactured by our industrial systems. Some people say that mobile phone manufacturers will evolve into robot manufacturers, just like our smartphones may evolve into small robots that can be carried around. Others believe that all electric car companies, including Tesla, will expand into robotics. Tesla has sent a clear signal – Musk has always emphasized that Tesla is a robotics company. In China, we have top electric car companies with strong export and manufacturing capabilities, all of which have the potential to transform into robotics companies. China has a big advantage here, not only in terms of manufacturing but also in terms of market size, especially given needs like senior care.
The second area of breakthrough is AI assistants or AI companions. They can assist us in our work, provide emotional value, and even become your romantic partner or personal doctor. Each of us will have multiple AI companions playing different roles. There are already many startups working with us to develop these types of products, starting with chat and role-playing apps. In a few years, these products will become smarter and gradually develop into assistants, AI doctors, companions, etc.
The third super application is autonomous driving. What I’m talking about is the collaborative construction of smart cars and smart roads, with AI organizing all transportation. China takes this very seriously and we are promoting connected cars and creating smart road infrastructure. Many car companies, including SenseTime, have made significant R&D investments in autonomous driving. Tesla's FSD is currently the world's leading autonomous driving technology and may also enter the Chinese market.
Moderator: Since artificial intelligence has such a profound impact on everyone, what role do you think artificial intelligence will play in solving inequalities in education, health care and other fields?
Xu Bing: I firmly believe that artificial intelligence should help reduce inequality. I think if the cost of AI dropped a million times, this should help the poor more than the rich. Now rich people can already get good medical services and children’s education services without artificial intelligence. But if the cost of AI drops to close to zero, we can put these services on everyone's mobile phone for free, which will help everyone, especially those in poverty, to get good medical care and education.
One of our goals is to provide excellent AI capabilities to the public for free. For example, many of the applications under SenseTime's RiRiXin model are free for everyone to use. We have even open sourced some of our best models.
Host: You have been in business for ten years and have faced many challenges along the way. How do you continue to stay motivated?
Xu Bing: This is a very good question. First of all, I think we're probably at the best starting point in history because we're experiencing what's probably the biggest technology wave in history. This means we can try many times in the right direction and have a greater chance of success. This will make us more adaptable, resilient and able to transform quickly.
I don’t consider any challenge or failure as a baggage. I think this is a very special era, and being able to be adaptable and act decisively and quickly can be greatly rewarded. So that’s how I stay motivated, not to dwell on setbacks or when things don’t go my way, because the next few years are going to be full of opportunities. I can’t imagine anything more exciting and inspiring than artificial intelligence.