Zuckerberg: I believe in AI, so at all costs, I invested hundreds of billions of dollars to build the strongest computing power and team

In the increasingly fierce competition in AI, Meta is redefining the future of super intelligence in a completely different way.

Recently, Meta founder and CEO Mark Zuckerberg revealed the company's new strategic layout in the field of artificial intelligence in an exclusive interview with The Information. Unlike competitors who focus on enterprise-level AI applications, Zuckerberg proposed a new vision of "personalized super intelligence," planning to put this technology directly into the hands of billions of users. At the same time, Meta is investing an unprecedented scale of capital in the AI field—hundreds of billions of dollars to build ultra-large computing clusters, with the largest Hyperion project covering an area almost comparable to Manhattan.

Zuckerberg stated in the interview: "The mission of our lab is to provide personalized super intelligence to everyone in the world, so we can put this power in everyone's hands. This is different from what other labs are doing."

Regarding the talent war, he said: "In the long run, if you plan to invest hundreds of billions in computing power and build multiple gigawatt-level clusters, it is indeed worth competing fiercely at all costs to secure those 50 or 70 top researchers to build your team."

Here are some key points from the interview:

Personalized Super Intelligence: Meta's Differentiated AI Vision. In the current AI race, most tech giants are focusing on enhancing enterprise productivity and economic automation. Zuckerberg stated, "So far, I think you hear a lot from other labs about wanting to automate all economic production work in society. Of course, I believe economic automation and economic progress have great potential to do a lot of great work that can help many people. But there is another side, which is what people care about in their own lives." Zuckerberg believes that while solving major problems is important, what people really care about is often relatively simple things in life. Meta's value has always been to try to put this power directly into people's hands, which is what it aims to achieve with "personalized super intelligence."

Gigawatt-Level Computing Clusters: Unprecedented AI Infrastructure Investment. Meta is building multiple gigawatt-level data center clusters. Zuckerberg believes the company has a very strong business model to support this. He revealed, "Our teams are working day and night on Prometheus and Hyperion, which are our first two Titan clusters, both of which will exceed 1 gigawatt. Hyperion will expand to 5 gigawatts in the coming years. I've shared pictures of it, but in terms of area, the scale of this site is comparable to a significant portion of Manhattan. It's enormous."

Talent War: Minimum Staff with Maximum GPUs Strategy. When asked whether they are really offering top talent compensation packages of $100 million to $200 million, Zuckerberg responded that many specific details reported are not accurate. However, it is indeed a very hot market. He stated, "If you are going to spend hundreds of billions on computing and build multiple gigawatt clusters, then it is really necessary to compete super fiercely and do whatever it takes to secure those 50 or 70 or any number of top researchers to build your team." People say, "I hope the number of people reporting to me is minimal, and the GPU is maximized."

AI Glasses: The Best Form of Future Interaction. Zuckerberg believes that AI glasses will be the best form of AI interaction and may even become a necessity for humans. He further explained the value of AI glasses: "In the future, we will have glasses that, if you wish, will be able to observe what happens in your life and follow up on things for you, either later or provide you with information in real-time through the glasses."

Capital Challenges from Competitors. When asked whether Meta's capital advantage poses a challenge to competitors, Zuckerberg answered affirmatively. He pointed out that Meta's strong cash flow capability allows it to fully self-fund these large-scale investments without needing to seek external financing like some competitors. "I think this is an advantage," Zuckerberg stated candidly. This capital strength is not only reflected in infrastructure development but also becomes an important factor in attracting top talent.

Future Outlook: The Arrival of the Superintelligent Era. There are differing opinions in the industry about when superintelligence can be achieved. Some believe it will take three years, while others think it will take five or seven years. Zuckerberg's attitude is quite positive. In an interview, he stated, "I think no one knows the answer. I just think we should bet and act as if it will be ready in the next two to three years. I believe there is such a possibility. If this is what you believe, you will invest hundreds of billions of dollars to build the strongest team."

Below is the transcript of the interview (translated by AI).

Jessica Lessin: Mark, thank you for joining The Information's TITV program. It's great to have you here. Just this morning, there was news that your significant announcement about the new supercomputer means you are investing more. We'll talk about that later. But first, you've made a huge investment in ScaleAI. You've been on an AI hiring blitz. Why? Why now?

Mark Zuckerberg: Yes, it's indeed very busy. You know, I think the most exciting thing this year is that we are starting to see early signs of models self-improving, which means that the development of superintelligence is just around the corner, and we just want to make sure we are doing everything we can to ramp up our investments and go all in. The mission of our lab is to provide personal superintelligence to everyone in the world so that we can put this power in everyone's hands. I'm very excited about this. It's different from what other labs are doing.

My view is that this will become the most important technology in our lives. It will underpin everything we develop in our company and will have a very broad impact on society. Therefore, we just want to ensure that the best talent is involved in this work, from entrepreneurs to researchers to engineers working on data and infrastructure.

Of course, we also want to support all of this with the absolute massive computing power we can provide because we have a very strong business model that can generate a lot of capital

Jessica Lessin: After the launch of Llama 4, do you feel like you're falling behind? It seems that especially this summer, you've slightly changed your strategy.

Mark Zuckerberg: I think this field is accelerating, you know, we are constantly setting the goals we want to achieve. And then the pace of development in this field always exceeds our expectations. So, you see, I'm not someone who is particularly satisfied with my position. I always want to do better. But what I'm really focused on here is the technology that may be developed in the near future. This will be super foundational, and I just want to make sure we are positioned correctly.

There is a big debate in the industry now: well, is superintelligence achievable in three, five, or seven years? But I think no one knows the answer. I just believe we should bet and act as if it will be ready in the next two to three years. I believe there is that possibility. If you believe that, you will invest hundreds of billions of dollars and build the strongest team at all costs.

Jessica Lessin: What about the reports that you are spending up to $100 million or $200 million on packages (compensation packages) to recruit talent? Can you comment on that? Are those numbers accurate? In this ongoing talent war, how do you view the price of talent?

Mark Zuckerberg: Well, you see, many of the specific details reported are not accurate. But it is indeed a very hot market. I mean, as you know, there is a small number of top researchers, and all different labs are competing for them. So I think the competition is definitely quite fierce.

What I want to say is, in the long run, if you plan to invest hundreds of billions of dollars in computing power and build multiple gigawatt-level clusters, then it is indeed worth competing at all costs to secure those 50 or 70 top researchers to build your team.

I think the reasoning here is that you don't need a large team to do this. You actually want the smallest team that can have a comprehensive understanding of the entire project. Therefore, there is definitely a premium for the best and most talented individuals. So, from this perspective, the amount spent on recruiting talent is actually still quite small compared to the overall investment in superintelligence, which makes sense.

Jessica Lessin: Obviously, you have competitors like Microsoft, OpenAI, Google, many of which have large enterprise businesses. How do you view Meta's opportunities? Is it around the same vision of productivity, or something else?

Mark Zuckerberg: I think personal superintelligence is a vision that is different from what other companies are pursuing. It's a bit like when the internet first scaled, people were asking whether the internet was for increasing productivity? For entertainment, or would it change the way we work in companies? The answer is all of the above. I think AI will be the same; different companies will focus on different parts. So far, I think you've heard a lot of other labs talking about wanting to automate all economically productive jobs in society

Of course, I believe that economic automation and economic progress have great potential to do great work that can be very helpful to many people. But there is another side, which is what people care about in their own lives? Part of it is productivity. But a large part is relationships, culture, creativity, play, and enjoying life. My guess is that many people in other labs are talking about, well, we want to develop this central superintelligence and use it to solve major problems.

I think the reality is that there are many great things to be done, and I am excited about it. And I think, when you put personal superintelligence in the hands of people, part of what they want to do is use it to solve major problems. But I actually think that many of the things people care about are just relatively simple things in their lives. And part of our spirit and values is always trying to put that power directly in the hands of people.

That is what we hope to achieve with personal superintelligence. So, you know, this is a bit of a different emphasis.

I think this will also shape how we build companies, how we write all the software, all the technology, and products. So this is likely to be very similar to many other places, and many of the fundamental approaches are similar. But I think the products we are going to build will be quite different, and I believe they will touch the lives of billions of people.

We are the leading company in developing products and scaling them to billions of users. Of course, we are also at the forefront of the next generation of computing platforms, and our glasses perform exceptionally well. I still believe that will be the best form of AI because, you know, they can see what you see, hear what you hear, and you can talk to them all day long. Once we add displays and holograms to them, it can generate a user interface (UI) for you. So I think there will be incredible connections between these things.

Jessica Lessin: So I’m imagining my Meta superintelligent glasses, companions, friends, doing things for me. Am I on the right track?

Mark Zuckerberg: Yes. I think we will reach a point where you are in this world and having conversations. One thought experiment I like to do is that in almost every conversation I have, there are about five things I want to follow up on, but I almost never follow up on all five, right?

Just because I go off to do other things, and if I’m lucky, I might follow up on one. And in the future, we will have glasses that, if you want, will be able to observe what is happening in your life and be able to follow up on things for you, either later follow up with you or give you information in real-time through the glasses. It’s almost like, if you need vision correction but don’t have glasses (optical glasses), you are at a cognitive disadvantage.

I think in the future, if you don’t have AI glasses, you will basically also be at a cognitive disadvantage. So yes, I mean, that’s part of what we plan to do. But we are also very focused on entertainment, culture, and that kind of personal relationship with people. I think, having an AI that can help you in this regard will be very valuable. ** So no matter what, there is a lot to be done. There is no shortage.

In addition to direct products, superintelligence will obviously change the way we develop in the company and the process of all our software development. I believe our engineers will be much more efficient than they are now. So I know this will be very exciting, and it is something we really want to focus on, especially the self-improvement loop.

We have already seen some signs of the model, even models based on Llama, autonomously improving Facebook's algorithms and other AI systems in the company. So, I mean, seeing these signs is quite exciting.

Jessica Lessin: I know you have to go, but I really want to ask about your news. But first, a different question, how do you name your data centers?

So, you like Titans, and now we have Hyperion, and so on.

Mark Zuckerberg: (laughs)

Jessica Lessin: Do you have limits on your capital expenditures?

Mark Zuckerberg: I think we will observe technology trends and see how the results turn out. When operating a company, I am always looking for ways to convert capital into higher quality services for users. One benefit of reinforcement learning is that it provides a place where you can potentially convert a very large amount of capital into increasingly better services, which may not be achievable by competitors with less funding or less boldness.

I think this is a competitive advantage. If we can get this mechanism to work well, that is also why we are basically going all out. We are building multiple gigawatt-level data centers, and we can basically support all of this with the company's cash flow. We have a very strong business model to support it. We have a very strong infrastructure team that is innovating to build the data centers.

You know, I hope they don't just spend four years building those concrete buildings. So we pioneered this new approach, and we are basically building these weatherproof tents and setting up networks and GPU clusters inside them to get them built faster. They are hurricane-level tents.

Our team is working day and night on Prometheus and Hyperion, which are our first two Titan clusters, both of which will exceed 1 gigawatt. Hyperion will expand to 5 gigawatts in the coming years. I have shared pictures of it, but in terms of footprint, the scale of this site is equivalent to a significant portion of Manhattan. That thing is enormous.

There is a lot of manpower and resources being invested in its construction. This is a good thing for the community. These are very exciting projects, and you can tell from the interview that I am very excited about building them. I am very excited because I believe we will have the largest computing power cluster in the company, driven by a lean and talent-dense team, and I believe we will have far more computing power per researcher than our peers for cutting-edge work I believe the next few years will be a very exciting period of significant development.

Jessica Lessin: One thing that impressed me about your announcement on Facebook is that you have the capital to do this. It makes me feel like this might be a slight jab at competitors who may still need to raise funds to build. Do you see this as an advantage?

Mark Zuckerberg: In my view, I think it is an advantage. I think it is an advantage. Obviously, I have spent a lot of time on recruitment, and there have been many reports about money and so on. As I mentioned before, many specific numbers are not accurate, but I think this underestimates other key reasons why people are so excited to work at the Meta Super Intelligence Lab.

One of the biggest reasons is that as a researcher, you can have greater leverage. You have more computing power, right? I mean, historically, when I recruited people for different departments in the company, people would say, "Well, what is my scope of responsibilities?"

Here, people would say, "I want the fewest people reporting to me and the most GPUs available." Therefore, having basically the highest per capita computing power is absolutely a strategic advantage, not only for getting the work done but also for attracting the best talent. I actually think this is one of the main reasons people are excited to do this, and it has been underestimated. Another reason is that having a clean slate is interesting in itself, right? I mean, many times people mention building infrastructure for these things.

Someone once gave me an analogy: it's like building a house. You build the first one for the enemy, the second one for friends, and the third one for yourself. I think being able to start from a clean slate, not working on someone else's infrastructure at another company, but having the opportunity to build a new lab from scratch is really appealing to many people. In short, yes, I am very focused on this.

Jessica Lessin: Your recruitment efforts are clearly successful now. I think this will be a long, long race worth watching. Oh, I have to ask. Did you meet Sam Altman at Sun Valley last week? After poaching so many people from OpenAI, if you did meet, what did you talk about?

Mark Zuckerberg: Well. I think part of the fun of the Sun Valley summit is that you get to meet all your friends and colleagues. So yes, I met Sam and said hello. I also saw Dario, Sundar, Satya, and everyone else working in this field.

Well, what can I say? We are building this new team from scratch, and there are probably only five or six places in the world with a lot of excellent researchers. Of course, Meta is one of them. Part of what we are doing is bringing many key figures into new projects.

But we are also conducting large-scale recruitment, so we are not targeting any specific person. I want to ensure that I personally know all the top researchers in the industry. That way, whether now or in the future, if there are good collaboration opportunities, we can facilitate them

Jessica Lessin: Well, it's a hot summer for AI, Mark, and you are at the center of it. Thank you for joining TITV. We will continue to follow up