Joe Tsai on the Future of AI

[00:00:12][music] >> Wow.

[00:00:18]What a big room.

[00:00:20]Bonjour.

[00:00:23]Are you enjoying today?

[00:00:26]Yeah?

[00:00:29]A bit sleepy.

[00:00:34]Okay, now you will have a great, great, great session.

[00:00:39]Joe Tsai from Alibaba with Christophe Jakubyszyn from Les Echos.

[00:00:46]As you know, VivaTech is a worldwide platform and we are pleased to welcome more than 170 countries. And when you walk through the alleys, you will see a a lot of pavilion from countries and a lot of nationalities.

[00:01:04]We hear a lot about our friends from the US, from Europe, of course, but too often we tend to forget the other other half of the world and that's why our next guest is here.

[00:01:23]He co-founded Alibaba.

[00:01:27]People do not know very well his name, but they will remember forever because he has helped transforming Alibaba.

[00:01:38]And Alibaba has created Quen.

[00:01:42]And I'm sure that you don't know, but Quen is the most widely used open source model in the world.

[00:01:54]The most widely open social AI model. It's not ChatGPT, it's not Claude, it is Qwen.

[00:02:08]And Alibaba released its latest version of Qwen a few months ago, and we are pleased to welcome someone who has been at the helm of the company since its inception.

[00:02:23]And he was the man behind the open-source of AI. So, please welcome very warmly Christophe Jakubyszyn from Le Wagon and a big round of applause for Joe Tsai, chairman and co-founder of Alibaba.

[00:02:43]>> [music] >> Joe, welcome. Christophe, the floor is yours.

[00:03:03]>> Hello everyone. [music] Thank you to be here. We are very, very pleased to as Maurice said to to welcome Joe Tsai, the co-founder and the chairman of Alibaba.

[00:03:12]Um Alibaba went through many revolution, and of course we'll speak about the last one about the open-source LLM model.

[00:03:20]Uh but first, can we talk about your history? Because um many people in France still feel things you are a marketplace, B2B, B2C. We know AI Express here.

[00:03:31]I know you are the largest marketplace in the in China. So, could you tell us more about your the evolution of your company, of your group?

[00:03:40]>> Yeah, thanks, Christophe. It's uh Maurice, it's great to be here.

[00:03:44]Uh so, Alibaba started in 1999 as a B2B marketplace. And at the time, the business idea was very simple. We put these small Chinese manufacturers and trading companies online to sell to the whole world wholesale. So, that kind of was on the cusp of of China entering the WTO. So, the export business was about to boom. And then we got into the consumer B2C business and launched a website called Taobao and which today is the largest e-consumer e-commerce business in China.

[00:04:20]>> How many customer?

[00:04:21]>> We serve 820 million Chinese consumers.

[00:04:26]And by the way, this platform helps European companies and brands to to sell 30 billion euros to Chinese consumers every year. So, you would say that Alibaba is quite important for European companies doing business around the world. But you what we are are went big into AI and cloud. And this is is actually it's quite a long history. We've been investing in cloud technology since 17 years ago. And that's because of necessity.

[00:05:00]We were seeing that we were managing so much data from our e-commerce business and transactions on a daily basis that we if we continue to be dependent on other people's technology like database and storage technology from others, we would end up giving all of our profits to the technology providers.

[00:05:23]And hence there was an effort to develop our own proprietary technology to manage all the data. All right? And and that's when our cloud business was born.

[00:05:33]Basically, we ate our own dog food before we launched the business to offer that service to third-party customers.

[00:05:41]But I think that that's sort of from the micro perspective from macro perspective, We're all in on AI.

[00:05:48]Uh it's actually the logic is very simple. Uh AI is today, if you ask me, how big is the market, what is the demand, what is the TAM, total addressable market, uh I would tell you that it's much, much bigger than any anybody's IT budgets. It's much, much bigger than the software market because what is AI? AI is producing units of human intelligence and human productivity. And if you look at global GDP over a hundred trillion dollars of GDP, uh at least half of that, 50 trillion, 50 trillion dollars is about human productivity and human intelligence. And that is the TAM of AI, and that's why we're all in on AI.

[00:06:32]>> So So you think AI will really um make us have a higher productivity because we we just had like 1 hour ago a round table where we were questioning that. Uh We people are investing a lot, but they don't see the results yet.

[00:06:47]>> They don't see the results yet. I think a a lot of corporate CEOs will tell you that their engineers are burning a lot of tokens, costs goes up. Uh but I think I would say we're at the cusp of uh real productivity gains. Uh right now a lot of people are experimenting. And there are uh definitely like we see it in our company, our own engineers. There are super users that uh use the coding tool, for example, and they are uh not just doing their work that that's within their domain, but they're also experimenting with other stuff because when you give engineers toys, they will always play uh with it more than and and they didn't realize uh that the company's actually paying for it, right? So this is what's happening right now, but I think what what I truly believe, this is really a belief of whether, you know, uh production of artificial units of intelligence will be able to add value to human intelligence and um it's like religion. I don't I don't want to convince all of you that that's going to happen, but we believe that that's going to happen.

[00:07:55]>> So if we go back to the Alibaba evolution in which layer of AI do you do you invest most? Is it infrastructure models, cloud services?

[00:08:05]>> So we are in at least four layers of AI.

[00:08:10]At the at the the bottom we're not involved in energy because in the China context energy is actually very efficient, less expensive, and the Chinese government has invested over the last 15 years in the national grid that makes energy delivery production and delivery very very efficient to all the users.

[00:08:32]But we are at the from the bottom we're involved in chips, infrastructure in terms of our cloud business, and also the model. We we have as Maurice very uh you know you know kindly pointed out that we our Quen model is actually one of the most popular open source models in the world.

[00:08:51]And then at the application layer we have an e-commerce ecosystem. We have an ecosystem for online shopping, grocery delivery, travel maps and all of that that can where we can infuse and deploy AI for our users.

[00:09:08]So we're involved in all those stacks of technology and we we think that there's huge benefits to the full stack approach as opposed to just taking one layer of the stack because I think over time nobody can tell you right now where the value will accrue.

[00:09:29]Which part of the stack? Is it in chips?

[00:09:31]Is it in cloud? Or is it in the model?

[00:09:33]Right now the model companies and pure model companies are very hot. They seem to accrue uh a lot of the value, uh but over time that may not be the case.

[00:09:43]Uh so, having a integrated approach of being involved in the full stack uh makes sense a lot of sense to us, and that's our core strategy.

[00:09:51]>> That's very interesting. You you right, we don't really know where the value will be, but uh when you see, you know, the the investment in infrastructure for AI, do do you think there is kind of bubble there? Is it Do Do we have the all those need to make the model work?

[00:10:04]Because some model are more efficient than others, and they they need less uh capacity than others. So, what about that?

[00:10:12]>> I don't think so. Uh I mean, the numbers really are quite astounding uh because if you just look at the American hyperscalers, the four or five companies combined will invest over $800 billion of CapEx, and next year it's going to go to over a trillion dollars. It it's it's eye-popping type of CapEx investment, and and I think it's natural that people will ask whether there's going to be overcapacity.

[00:10:39]Uh but again, like I said, we're going against We're We're trying to tackle a total addressable market of $50 trillion US dollars. And that's uh that's why we're optimistic. And in the China context, uh we're actually very underinvested in uh infrastructure and in the supply chain of AI. Uh and so, it actually behooves all the companies in China to step up their investment. Uh obviously, we're not investing at the same level as some of the American hyperscalers, but it is still very uh uh very substantial.

[00:11:16]>> Why don't you?

[00:11:18]>> Oh, well, you know, sometimes you're limited by uh the your your capital. Uh you're limited by the free cash flow that you're generating. Uh and the good lucky thing is Alibaba is one of the very few companies that actually has a core business, which it is e-commerce.

[00:11:33]We generate $25 billion of free cash flow every year from our e-commerce business that can fuel our investments in AI. So, we're we're actually one of the better positioned companies uh out there.

[00:11:45]>> I think today your your your your business in in um marketplace is still 80 85% of your of your turnover.

[00:11:52]>> Yeah, it it is still 80 plus percent of our revenues in the e-commerce marketplace.

[00:11:58]Uh uh we're very lucky that that produces the cash flow that allows us to make the future investment.

[00:12:04]>> in AI and cloud business.

[00:12:05]>> Yes, absolutely.

[00:12:06]>> Uh what about your your Maurice mentioned in the the the Qwen uh offer and the Qwen model? What what what because it's open source now, so so >> Yeah.

[00:12:14]>> what what kind of customer and you you can have and you can help with that?

[00:12:19]>> Well, I thought Maurice was too charitable in crediting me to as the open behind the open source movement. That that that's totally untrue. Uh I would say that our colleagues at Alibaba have worked last several years in pushing uh the open source movement in uh from these frontier models and to to a high degree of success. And we're continuing to do that.

[00:12:45]Um what's the significance of open source?

[00:12:49]So, I've been spending the last couple of weeks in Europe. And as I talked to European companies and CEOs and uh scientists, people in Europe, one of the biggest key words here is sovereignty.

[00:13:02]Uh but what is sovereignty? If you ask 10 Europeans about what sovereignty means, you'll get 12 different answers.

[00:13:11]And um uh but to me, they mean basically two things. One is technology independence.

[00:13:19]Uh they they are all worried about the kill switch. Uh that being relying on some other countries' technology, uh they could turn shut off the kill switch. We just saw a very live example of that in the recent days.

[00:13:31]Yeah. That's right. And the the second is data privacy. People wanted to be able to use AI, use technology with data that's proprietary to them within their own environment and hence build up the the firewall to protect their own data. Uh, I think open source solves both problems because it is basically a free piece of software that you can download into your data center.

[00:13:59]Uh, you can develop you can download the Quen model in your own notebook computer, for example.

[00:14:04]Uh, and that that's completely independent from uh, the original maker.

[00:14:10]Uh, so if you use our open source Quen open source model, it will have nothing to do with Alibaba. I mean, we have to figure out how we charge for it, but we can't. Um, so that's number one. That's independence.

[00:14:22]What's important is you can take the open source model and then you could take your own data and train it further train it, fine-tune it, do post-training of the model. Uh, and and keep all your entire data private within your own firewalls. And I think that's a really really important point uh, for companies in Europe to recognize that uh, open source is actually one I'm not saying it's the panacea. It's not the only approach.

[00:14:50]It's one of the approaches to achieve some degree of sovereignty. And uh, today uh, the open source movement is actually driven by Chinese companies.

[00:14:59]Uh, all the American players uh, have a closed source uh, their their model. So they want you to use uh, their models through an API and you have no idea where your data is going.

[00:15:11]When you when you're chatting with the uh, you know, a chatbot, uh, all of your most private questions and your confessions uh, go into some uh, you know, some pool where they use that to further train their model. So, you have no idea where that that's going.

[00:15:28]>> But Troy, to be honest, we we you know, Europe sovereignty is a very great concern today and we I think we just realized how dependent we can be from the US or from China from the industry for instance. So, um how I mean I I I agree with the open market I mean the the open model etc. but but still uh we we we can be worried there is you know, bypass for the Chinese company to maybe to one day they want to cut the access or to to be able to stop the the model to function. How how can how can we trust China as as we because we were disappointed by the US in the last days but but we could be also by China. I mean it's it's a big risk for Europe.

[00:16:05]>> Yeah.

[00:16:06]You can't.

[00:16:08]That's the short answer because you cannot be uh uh uh relying on a a third party government to say that they won't do something that's detrimental to you. But here's the thing, right now all of your eggs are in one basket. Why not get a second basket? Uh to put your eggs in two baskets even though Europe at in in the long term may uh develop their own basket, but at least now you have two eggs uh sorry, two baskets to put your eggs in.

[00:16:38]>> Yeah.

[00:16:39]>> That's true.

[00:16:40]>> Uh uh >> May uh maybe we can you you can give example of how you work with the German company like BMW. You you you you help them and or Siemens in Germany. Uh what what was the base of the collaboration and cooperation you did with them?

[00:16:59]>> Yeah. So, these German manufacturing businesses are absolutely fascinating.

[00:17:04]They're all clients of ours, Alibaba Cloud in China uh in their Chinese operations. Uh and we work with them in in the manufacturing contest uh in terms of design, uh, testing, quality control.

[00:17:21]And, uh, uh, we think that, uh, this is in the future is going to be a very interesting segment because today most of the application in AI, you know, you have ChatGPT, that's consumer. You have like Claude Co, that's, uh, uh, focused on, uh, coding and the knowledge worker space.

[00:17:40]Uh, but in the future uh, these manufacturing businesses are very valuable because they will have very valuable data that's proprietary to them in the manufacturing process. And that data is it is very, very good, high-quality data that you can use to train your own model uh, to improve your manufacturing process. And this is the approach that we're taking uh, with these companies. Uh, you you also, uh, you mentioned BMW, uh, Siemens. Uh, we also have worked with Bosch. Last week I was at the Bosch Connect World Conference. Uh, we work with them because they are, uh, using AI uh, to develop their, uh, assisted driving, autonomous driving.

[00:18:23]And, uh, that requires a lot of compute.

[00:18:26]Uh, uh, so, you know, uh, there there's a lot of very interesting going on in the manufacturing sector.

[00:18:34]>> So, if I understand you you consider the ban on, um, Anthropic last model by the US an opportunity for you and your model to be, uh, to be adopted by, uh, other European customers?

[00:18:44]>> Our model?

[00:18:45]>> Yeah, so, your model, yeah.

[00:18:47]>> Uh, yeah, well, so, there's two approaches to it. Uh, one is, uh, through, you know, they could just take our open-source model and then, uh, they will put that, deploy that into their own infrastructure if they have their own data center, whatever.

[00:19:01]And, uh, uh, but our infrastructure is developed hand, uh, sort of, you know, hand hand with the model. So, we actually have one of the most efficient infra to help people train their models.

[00:19:17]And uh so if they use our open source model, they could also buy compute from us, right? So that's a symbiotic very symbiotic relationship between the model and infra.

[00:19:27]Um So uh that's one approach. Another approach is there are now emerging uh uh a num- a number of uh companies that uh develop uh these uh inferencing platforms uh that offer people a choice of models. Uh different kind you don't have to use one, you could use somebody else's. Uh they also can access closed source models as long as there's an agreement between the model maker and the inferencing platform to open up the weights uh the gates to the weights uh in a private context. Uh and then customers can go on the go on to these inferencing platforms to use the models.

[00:20:08]>> I have a more difficult question about um the your vision of future of AI LLM balance between humans and and agents uh and even in humanity. How do you see the the the next 10 years of the humanity?

[00:20:24]>> Uh so today I was um I had a a talk with uh the Alibaba Paris office colleagues.

[00:20:33]Uh it's located we just moved to a new office. It's located in the I think the second or third floor of a beautiful building.

[00:20:42]And I looked out the window, there's a cafe.

[00:20:45]And people are sitting in the cafe uh outdoors because the weather is pretty nice having a great time.

[00:20:55]And I point to them and I tell my colleagues, this is the future of AI.

[00:20:59]They may be you think they may be drinking cat coffee and you know, having a good time, not doing anything, not being productive, but the reality is uh they have deployed agents that are doing the work. So, when you're going to sleep, you still have agents working for you.

[00:21:18]So, think about the productivity gains.

[00:21:20]If 24/7, you can actually have somebody else work for you.

[00:21:24]>> So, you have you have you have the same philosophy that the the the people from the Silicon Valley, you think that many people won't work and we will make the the agent and robot work for them?

[00:21:34]>> Well, I I think it'll free definitely free up people's time to do to enjoy life, to enjoy their family, to do more entertainment. This is why I'm very big on live entertainment. I mean, when people spend less time in the office, where do they want to go? They don't want to just sit at home. They want to go to concerts, they want to go to football matches, they want to go to basketball games.

[00:21:57]That's why >> that's to French people because they will they will want to work less.

[00:22:01]>> [laughter] >> Well, a lot of them are focused on the World Cup right now. So, >> Chinese worker will will work a lot when you when you when you look at the Chinese engineer, they they have lots of working hours. Even with agent, even with AI.

[00:22:16]>> Uh there will always be people that will work a lot more than others, but I think most of us want to be able to enjoy life a little bit, want want to be able to spend time with our family more.

[00:22:27]>> I think you are you are you are very big a big fan of the basket Brooklyn Nets and even the Liberty New York supporter. I think you are even support a player I even don't know Lacrosse.

[00:22:40]Uh and what about soccer World Cup? Are you are you enjoying the the the soccer?

[00:22:46]>> Uh first of all, about basketball, I went to the French League Championship game last night between Paris and Monaco. Uh this is a second game of the five-game series.

[00:22:58]Uh and I thought it was very interesting. And then I realized that uh our women's basketball team, the New York Liberty, have three French players.

[00:23:06]Three players that play for the French national team. And then we have on the men's Brooklyn Nets have one French player. So there's a lot of very big French connection.

[00:23:15]I think France is probably the most important non-American influence in the NBA and also in the WNBA today. So it's very interesting.

[00:23:27]So about the World Cup? I I don't have a horse in this race.

[00:23:34]That's the problem.

[00:23:36]I think I just saw some social media that says there's a Chinese referee in the World Cup and he is getting sponsorships. And I hope at some point China will have a strong team, but I think that development will take some time. Yeah.

[00:23:53]>> Thank you very much, Joey. Thank you.

[00:23:54]Thanks to you.