allows us to see things that, you know, the frailty of the human condition, we can't quite see yet today. And I think we should just be open-minded and prepared to deal with that as a community because I think it's going to be complicated. So there's no bubble in that context.
Yeah. Chamath just raised a very interesting point which is that there is no bubble. If you take these models operate on the current path of trajectory and if there's some breakthrough change then the underlying energy infrastructure, all that stuff is going to that becomes a bubble in itself. It's sort of a bubble that happens because conditions have changed. That's a I think that's I wouldn't call it a corner case but that's something that people should think about. But I think if you look at right now the massive amount of capex investment that all the hyperscalers, all the model companies are making. I mean the latest quarter reports that people are doubling their capex. It used to be last year these hyperscalers are making 60 to 80 billion per company per year of capex and now it's doubled like 120 to 150 billion per year. It's astounding just within a year right and but here's the thing I think a few things are driving it. Number one people still believe in the scaling law. And now also the idea that inference time scaling is important because more compute resources will be allocated to inference and inference is where you know everybody that uses AI will touch inference as opposed to the few model companies that only do training. Right? So a lot of capex that's going toward inference. And then finally you have people making multimodal models that deal in generative rich media, videos, images, that will burn a lot of GPU resources. So people seeing this scaling law and multimodal world and we haven't even talked about world models yet and that's the validation that there will need to be more compute that need to be invested in. I don't think these trends are going to reverse or veer off in the future. People will still continue to, you know, for example are people going to consume less videos? I don't think so. And so in that sense there's no bubble but having said that people are investing anticipating future demand as opposed to I think there could be you know in certain sectors you see you know like memory you see shortages today but the just the massive amounts of investment it's kind of hard to think about ROI. I hate to be the CFO of one of these companies because they have to they're faced of, you know, having to explain ROI to investors. They can't do that.
The theme for 2026 for those of you who've gone to Davos and for some of the other conferences and some of the other discussions is AI agents and you know agents are the name of the game where people have to actually try to work with agents, deploy, develop and utilize the different agents that are going to do different tasks. We have a number of heads of state and heads of government in the room. The question that you're going to get when you say that is we haven't started with deploying AI already and you guys are now moving into agents. So if someone wants to leapfrog someone who hasn't done anything yet, whether it's a nation in Africa or a nation in Latin America or a nation in Southeast Asia, what would your advice be? Should they go into the generative AI model side of things? Should they try to do a specific thing with agents? What would be your advice if you were advising them?
Neither. I think it's you have to pick a model and I think you need to deploy it inside your own borders with your own power and roughly your own people and be able to use it for a handful of very useful and practical solutions to everyday problems whether it's public health or whether it's governmental services or frankly the most important and valuable one is education which is being able to create a system that teaches the young people in your country how to learn in this new world where this supercomputer will be available to everybody. All this other stuff I think is very speculative. It's exactly what Joe said. Every frontier has a new language that it uses to justify further investment. And I think today because there isn't a super clear ROI, agents are the next vernacular that we've chosen to justify the next two or 300 billion dollars of investment. The more practical solution, I think, is to be one generation and one step behind that and say, let's go and educate our kids, cure some disease, and make it easier to renew our driver's license. And if you just do that, I think you're going to have a happy population.
So quality of life improvement.
Joe, I totally agree. I think governments have to think about what is the application where the use of AI will be productive and useful in education, in healthcare and you start you should start with the use cases first. What is the right application for AI? I think the idea of building up the full stack infrastructure that everything is a game for large economies. It's like building up a military industry. It's not for everybody. So once you've identified the use case and the applications then the question is all right whose technology are we going to leverage on? And on that front I believe that open source is the best way is the most secure way where you can have most control of your own AI.
Gentlemen, that was a great discussion. I know that we could have gone on for so much longer, but thank you for your insights and thank you for being here today. Thank you.
create for example a ultra high storage battery or that can effectuate a room temperature superconductor. If any of those things come to pass, we all would collectively have to underwrite what the actual value of our natural resources collectively as a world means. The change it would change how we use oil it would change how we use natural gas it would change how we build SMRs. So I think the honest question answer is this is why it's such a profoundly interesting moment all of the things that we've used to underwrite productivity and GDP will get requested because this super intelligence allows us to see things that you know the frailty of the human condition we can't quite see at today and I think we should just be open-minded and prepared to deal with that as a community because I think it's going to be complicated. So there's no bubble in that context.
Yeah. Chamath just raised a very interesting point which is that there is no bubble. If you take these models operate on the current path of trajectory and if there's some breakthrough change then the underlying energy infrastructure, all that stuff is going to that becomes a bubble in itself. It's sort of a bubble that happens because conditions have changed. That's a I think that's I wouldn't call it a corner case but that's something that people should think about. But I think if you look at right now the massive amount of capex investment that all the hyperscalers, all the model companies are making. I mean the latest quarter reports that people are doubling their capex. It used to be last year these hyperscalers are making 60 to 80 billion per company per year of capex and now it's doubled like 120 to 150 billion per year. It's astounding just within a year right and but here's the thing I think a few things are driving it. Number one people still believe in the scaling law. And now also the idea that inference time scaling is important because more compute resources will be allocated to inference and inference is where you know everybody that uses AI will touch inference as opposed to the few model companies that only do training, right? So a lot of capex that's going toward inference. And then finally, you have people making multimodal models that deal in generative rich media, videos, images, that will burn a lot of GPU resources. So people seeing this scaling law and multimodal world and we haven't even talked about world models yet and that's the validation that there will need to be more compute that need to be invested in. I don't think these trends are going to reverse or veer off in the future. People will still continue to, you know, for example are people going to consume less videos? I don't think so. And so in that sense there is no bubble but having said that people are investing anticipating future demand as opposed to or I think there could be you know in certain sectors you see you know like memory you see shortages today but the just the massive amounts of investment it's kind of hard to think about ROI. I hate to be the CFO of one of these companies because they have to they're faced of, you know, having to explain ROI to investors. They can't do that.
The theme for 2026 for those of you who've gone to Davos and for some of the other conferences and some of the other discussions is AI agents and you know agents are the name of the game. People have to actually try to work with agents, deploy, develop and utilize the different agents that are going to do different tasks. We have a number of heads of state and heads of government in the room. The question that you're going to get when you say that is we haven't started with deploying AI already and you guys are now moving into agents.
Gentlemen, it's an absolute pleasure joining you on stage today. I'd make a quick introduction. Chamath is a legendary investor. He runs Social Capital and he's recently done very well with Grok. Your investments there I think had paid off dividends but well done on that. Joe is a friend and the chairman of Alibaba. Last year on the stage we were talking about AI and where it was going and I must say all your predictions came true with regards to how fast China is moving when it comes to artificial intelligence and as well the ability of companies to iterate to innovate and to move forward. I'd like to start off by maybe talking about an interesting shift. We've always been used to seeing the US being the champion of open source, Silicon Valley leading that charge and the rest of the world focusing on trying to have their own advantage and focus on building closed source systems. With AI, we're seeing a complete shift. China's championing open source. The US is trying to push for open source but are working on closed source systems. How do you think that's going to affect the way that governments pivot, the way that people choose the AI models that they use? We'll start with Chamath and then we're going to go to Joe.
Well, I'm actually quite surprised by it as well. I mean, I do think that there are a handful of American open source models that are quite good. I think Nvidia's models are excellent. But in fairness, the Chinese open source models are just superb. So I think what it means is that every country probably in the next 3 to 5 years will be forced to make a very important decision which is around the sovereignty of their own productivity and GDP because to the extent that these AI and in that realm I think the open source models provide the clearest path because it just gives you total transparency into what's happening underneath the hood. Now there's a lot of work and we can talk about this later that has to happen for a country to be able to embrace an open source model. But I think the future is open source.
Well, I totally agree with Chamath what Chamath said about different countries you're better off with open source model if you are focused on sovereignty and having control of your own data. But just going back to history a little bit why China has evolved and why most companies in China have taken an open source approach is that software has in terms of application software like SaaS industry has never developed in China. So the idea of users coming to an API to use an AI model and pay for it has never been a primary business model and it's not and the judgment by all the model developers is that they're not going to make a lot of money providing a closed source through an API. So all the model companies have decided that why don't we just open source first and let the model take care of itself in terms of adoption of usage and proliferation. And fortunately a lot of various governments have taken notice and because they care about the sovereignty aspect and care about their data privacy. You could take an open source model and deploy it on your own infrastructure. You can do it on premise. You can do it in a virtual private secured environment. And whatever you do to post-train the model or running inference will have nothing to do with the original open source model maker and thereby you basically claim sovereignty and claim ownership of the model. But it remains to be seen. It's still a very big question how economically all the model companies are going to make it sort of sustainable with an open source approach.
Don't you think the issue is the economics? You're spending billions of dollars to build this model and if you open source them you're not really monetizing it in the way that you can potentially monetize it. So that's why most of the companies are focusing on closed source to try to generate as much profit from the model and then they open source the older versions. We've seen that with Grok for example we've seen that with even Llama and other models in the US where the frontier ones are closed source until they're able to generate some revenue or at least some advantage whether it's getting market share or otherwise. Do you think that is the main reason or is there more of an ideological shift and can China sustain making all these models open source especially the frontier ones?
So I would make a distinction here between companies that develop the frontier models plus an infrastructure business like cloud computing versus those companies that actually just only do the model. I think the companies that only do the model will have to think about how do you generate economics from open sourcing. The companies that actually have a cloud infrastructure for example in the United States you have Google that has a full stack of the model and the cloud infrastructure they also have chips can monetize in other ways from the usage of the open source model. Obviously Gemini is not an open source but if you look at China look at the example of Alibaba we have our Qwen model is open source but we also run our cloud business and today we're monetizing because people are training and also developing their models running inference on their models on our infrastructure and thereby generating the economics. But I think it remains to be seen how in the future how can you get especially consumers pay for a monthly subscription and generate economics there. That remains to be seen. That's the biggest challenge in the Chinese market.
I mean, I think the reality is that when you adopt an open-source model, you're going to need, as Joe said, all of these other things. You're going to need power, you're going to need infrastructure, depending on the solution that you use. You're going to need a lot of natural resources, specialty chemicals, you're going to need battery systems, and then you're going to need all the knowhow, and then on top of all that, you're going to need the capital and the financing. And so I think the reality is that the business model becomes more at the governmental level almost because if you take an open source alternative, but then the infrastructure of the government can come in and help you with all of these other things, you're going to take it. And so I think that is yet to play out but I think that's the real opportunity. Similarly in the United States it's all private. So the alternative is you know you can't really take a spin of Grok or take a spin of Gemini and implement it on your own for the needs of your own population. But yet you still have to bear the cost of all of these other things. So if all of this was packaged and financed and turnkey and in a box, I think that's the ultimate best solution. That's probably where international finance organizations and development banks probably need to play more of an evolving role over these next 10, 20 years.
Many of the conversations that I currently have and interviews that I would do and I'm sure that you guys will go through the same thing. People talk about the AI bubble and the bubble in a sense of is it sustainable? Can we keep seeing the numbers that we're seeing? We've recently seen the issue with memory and memory availability and we're not being able to produce memory enough. And when it comes to bubbles I think there are two types of bubbles. There's the tulip mania bubble which was completely made up. It blew up. People lost their money and there was nothing left except the legacy of the bubble. And then there's the railway bubble of the UK of the 19th century where people went and invested in railway tracks which gave the UK the legacy of being one of the most connected places on earth through railway. So there was actually a tangible legacy that was left behind. In terms of tangible legacies for artificial intelligence if and I'm not saying it is a bubble but let's assume that there was a set of bubbles along the way of the progress of artificial intelligence do you think that energy infrastructure is the safe bet? Is the safe bet investing in energy infrastructure? Is the safe bet investing in chips?
It's I think it's very much a TBD and the reason is that the capability of these models can actually disrupt the underlying asset that you may think is safe. So the example that I would give you is we are one or two generations of models away from a model demonstrating that there's a certain chemistry that can effectuate for example a ultra high storage battery or that can effectuate a room temperature superconductor. If any of those things come to pass we all would collectively have to underwrite what the actual value of our natural resources collectively as a world means. The change it would change how we use oil it would change how we use natural gas it would change how we build SMRs. So I think the honest question answer is this is why it's such a profoundly interesting moment all of the things that we've used to underwrite productivity and GDP will get re
Gentlemen, it's an absolute pleasure joining you on stage today. I'd make a quick introduction. Chamath is a legendary investor. He runs Social Capital and he's recently done very well with Grok. Your investments there I think had paid off dividends but well done on that. Joe is a friend and the chairman of Alibaba. Last year on the stage we were talking about AI and where it was going and I must say all your predictions came true with regards to how fast China is moving when it comes to artificial intelligence and as well the ability of companies to iterate to innovate and to move forward. I'd like to start off by maybe talking about an interesting shift. We've always been used to seeing the US being the champion of open source, Silicon Valley leading that charge and the rest of the world focusing on trying to have their own advantage and focus on building closed source systems. With AI, we're seeing a complete shift. China's championing open source. The US is trying to push for open source but are working on closed source systems. How do you think that's going to affect the way that governments pivot, the way that people choose the AI models that they use? We'll start with Chamath and then we're going to go to Joe.
Well, I'm actually quite surprised by it as well. I mean, I do think that there are a handful of American open source models that are quite good. I think Nvidia's models are excellent. But in fairness, the Chinese open source models are just superb. So I think what it means is that every country probably in the next 3 to 5 years will be forced to make a very important decision which is around the sovereignty of their own productivity and GDP because to the extent that these AI and in that realm I think the open source models provide the clearest path because it just gives you total transparency into what's happening underneath the hood. Now there's a lot of work and we can talk about this later that has to happen for a country to be able to embrace an open source model. But I think the future is open source.
Well, I totally agree with Chamath what Chamath said about different countries you're better off with open source model if you are focused on sovereignty and having control of your own data. But just going back to history a little bit why China has evolved and why most companies in China have taken an open source approach is that software has in terms of application software like SaaS industry has never developed in China. So the idea of users coming to an API to use an AI model and pay for it has never been a primary business model and it's not and the judgment by all the model developers is that they're not going to make a lot of money providing a closed source through an API. So all the model companies have decided that why don't we just open source first and let the model take care of itself in terms of adoption of usage and proliferation. And fortunately a lot of various governments have taken notice and because they care about the sovereignty aspect and care about their data privacy. You could take an open source model and deploy it on your own infrastructure. You can do it on premise. You can do it in a virtual private secured environment. And whatever you do to post-train the model or running inference will have nothing to do with the original open source model maker and thereby you basically claim sovereignty and claim ownership of the model. But it remains to be seen. It's still a very big question how economically all the model companies are going to make it sort of sustainable with an open source approach.
Don't you think the issue is the economics? You're spending billions of dollars to build this model and if you open source them you're not really monetizing it in the way that you can potentially monetize it. So that's why most of the companies are focusing on closed source to try to generate as much profit from the model and then they open source the older versions. We've seen that with Grok for example we've seen that with even Llama and other models in the US where the frontier ones are closed source until they're able to generate some revenue or at least some advantage whether it's getting market share or otherwise. Do you think that is the main reason or is there more of an ideological shift and can China sustain making all these models open source especially the frontier ones?
So I would make a distinction here between companies that develop the frontier models plus an infrastructure business like cloud computing versus those companies that actually just only do the model. I think the companies that only do the model will have to think about how do you generate economics from open sourcing. The companies that actually have a cloud infrastructure for example in the United States you have Google that has a full stack of the model and the cloud infrastructure they also have chips can monetize in other ways from the usage of the open source model. Obviously Gemini is not an open source but if you look at China look at the example of Alibaba we have our Qwen model is open source but we also run our cloud business and today we're monetizing because people are training and also developing their models running inference on their models on our infrastructure and thereby generating the economics. But I think it remains to be seen how in the future how can you get especially consumers pay for a monthly subscription and generate economics there. That remains to be seen. That's the biggest challenge in the Chinese market.
