Skip to content
Meta’s Secret AI Weapon: Its Data Centers episode artwork

Jul 20, 2025 12 min

Meta’s Secret AI Weapon: Its Data Centers

from Hard Fork AI

Meta’s Secret AI Weapon: Its Data Centers is focused on solving compute, storage, and energy challenges simultaneously. Our experts analyze the tradeoffs, benefits, and future of this direction. Try AI Box: ⁠⁠https://aibox.ai/ AI Chat YouTube Channel: https://www.youtube.com/@JaedenSchafer Join my AI Hustle Community: https://www.skool.com/aihustle/about

Episode details come from the podcast's RSS feed.

Embed this episode

Ready to play

Meta’s Secret AI Weapon: Its Data Centers

0:00 12:12
of matches

Transcript Auto-generated, may contain errors

Mark Zuckerberg has taken two threads to make a massive announcement. And that is that Meta is currently building a five gigawatt AI data center. Absolutely enormous. And there's some crazy news about them possibly just throwing up tents because they're trying to get this thing built so fast.

Mark Zuckerberg, who has famously been, it felt like, lagging a little bit on AI, does not like that, does not like the people that he was getting behind, and is now going full till. I feel like we kind of have these hot and cold cycles from Mark where he just goes really hard for a few months, then he chills while everyone does something, and then if anything, any crap starts in the fan or Meta starts getting behind in the area, he comes back full swing and goes crazy. So we're in one of those moments. He's been on an absolute tear, $100 million offers to tons of AI researchers.

He's just recruited tons of top talent from every single one of the top AI companies for his super intelligence project that he's building. And now the data center element is coming, which is going to be very interesting and allowing to compete with a lot of other big companies, including OpenAI, Google, and XAI in kind of this pairing that we're seeing where these software tech companies are not just building the models, but they're actually trying to build up the infrastructure as well because they know it's really just a very cutthroat race, especially when you have companies like Google that control data centers or Google Cloud and the AI models, and now you're trying to compete with Google. Maybe they are so desperate they're going to go to Google for some of the compute, and all of a sudden it's like you're funding your competitors going to turn around and use that money against you. So these guys really want to build it out on their own.

So we're going to dive into everything that he's announced, where we think this is going in the future. Before I do, I wanted to mention, if you want to try out all of the latest AI models, including everything from Meta, go check out AIbox.ai. That's my own startup that currently is in beta. We have a playground that allows you to use the top 40 AI models all in one platform for 20 bucks a month, so you don't have to have $20 subscriptions to all of these different companies.

We have tons of different open source models from Llama, including Llama 3.1, 405B Turbo, a bunch of other interesting models. We have Google, DeepSea, Cohere, Anthropic, OpenAI, XAI, Quen, NVIDIA, a bunch of cool models, and also a bunch of image models you may not have tried, and speak models. You can use them all in the same platform. In the same chat, you can switch between what model you're talking to.

So it's very useful. Go check it out, AIbox.ai. Okay, let's get into the latest news from Meta and Mark Zuckerberg. So this came out on Monday, but essentially this is the newest move that he's trying to use to get ahead of OpenAI and Google.

It's five gigawatt of computational power, essentially, for his AI lab. We know that he's got a bunch of players that are helping him leave the software side. He just was able to poach the former CEO of Safe Superintelligence. That was a company that Ilya Saskiver, who was one of the co-founders of OpenAI, he left, started his company, raised a billion dollars before a product ever existed.

Everyone's got this eye-watering number. And he had this incredible CEO who'd come over as well, which was Danny Gross. Now, Danny Gross is left, he's over with Zuckerberg now. We also have the CEO of ScaleAI, Alexander Wang, who also had a bunch of data, essentially, to help train AI models and had OpenAI and Google as a client.

And now Meta stuck him up with Google and said, hey, we're not going to use ScaleAI anymore because of kind of the way this is going down. But in any case, he's got this really kind of the dream team, this powerhouse team of people behind it. So what he needs now is the infrastructure. He's now turning all of his attention, now he's got an amazing team, to building out this massive computational, all the massive computational powers that we need to train the actual models that they are building.

So he announced the name of this, you know, this huge thing as it's called Hyperion. And he said that the footprint, the actual size of it is going to be big enough that it could cover most of Manhattan. So this is absolutely insane size-wise. Spokesperson for Meta, Ashley Gabriel, said that it's going to be located in Louisiana, probably in Richland Parish, where Meta already announced this $10 billion data center development they're going to be doing there.

And so it seems like this is probably going to be in the same place. They're going to bring two gigawatts of data center capacity online by 2030. So this isn't quite great. That's five years in the future.

Still a lot of compute. So in five years, they're going to have two gigawatts online, but they're going to scale it to five gigawatts several years later. So the plan is, you know, a lot of that is groundwork, and then once that's kind of built out a few years later, they'll be able to get it. So you can imagine by like 20, let's say 2033, maybe things get a little delayed, 2035.

It's like 10 years in the future before we're getting this thing like really humming at full capacity. So honestly, I think this is, it's kind of an interesting thing. It just shows how much work and how much time it takes to do this. It's honestly one of the reasons why a lot of people are very impressed by XAI being able to spin up a data center as fast as they did with 100,000 GPUs, is because this just takes a long time.

All of the, you know, getting everything done right with the government, with, you know, getting all of the right documents and everything in place to have this thing go through. It is quite a process. So Zuckerberg also said that Meta is going to bring a one gigawatt supercluster called Prometheus online by 2026. So this is much quicker.

Next year, at some point, we're going to get one gigawatt online. And so he's going to be one of the first tech companies to have an AI data center that's this big. So Prometheus is going to be located in New Albany, Ohio. Meta's AI data center build out is probably going to make them very competitive with opening iGoogle, Anthropic, and kind of what they're actually able to train and do.

You know, you kind of see one of two strategies where these AI companies are building out like Meta or Google, you know, the data centers. Or you see the strategy of something like Anthropic, where they're partnering with AWS, who already has a lot of infrastructure placed to use their infrastructure and go and do it themselves. So overall, you have a couple different options. Meta obviously is not going to rely on any other tech companies.

They know that they're quite literally competing directly with many of them. So between the two of them, what a lot of people are talking about, I don't know, it's not really my favorite talking point on all of this. I tend to think we're going to solve a lot of the energy problems. Like we know we have to figure them out, but there is the energy issue, right?

So we're building all these data centers. They're going to use a lot of energy. It's again, another interesting thing that we found with XAI when they spun out their data center very quickly, it's like where they get energy. There is a way to do this quite quickly.

I don't think that's going to be Meta's strategy, so they're going to have to figure that out. People also have kind of accused them of like, oh, it's going to be bad for electricity or it's going to be bad for water because it's going to use a lot of water. By one point, I'll make up water. And I think energy is a legitimate concern.

We've got to figure out what we're doing. And by what we're doing, we've got to figure out how to build more energy compute in America. That's probably nuclear. That's also something that takes a long time to build up.

China's building up all the energy capacity that they need. They're going to completely smoke us in AI if we don't build it up. So I think the energy problem is not one to take lightly. It's very serious.

And we've got to figure out how to get more energy creations out of America. The water one concerns me a lot less. They talk about, I don't know, apparently in the New York Times, they said that in Newton County, Georgia, Meta's data center project they have over there has made it so that the taps have run dry for some people's homes. I don't know what exactly the issue is there.

I don't imagine this is a long-term thing where people's taps in their house are just going to run out of water because the data center is soaked up so much water. You'd imagine you probably want to build these things in places where you have easy access to a lot of water. So in any case, I'm not sure what the problem is there. Also, the other thing that I would love to bring up is just the fact that when we talk about data centers are using water, it's not like they take in like infinite water and then just like use it and it disappears.

What the water is used for is cooling the machines down. So it's actually circulating in a reservoir and recycling. So yes, like maybe when Meta is building this data center in Georgia, maybe filling up the reservoirs takes a lot of time, but it's like the water all turns into steam and disappears. They keep it inside of the system.

They're using it for cooling. They're recycling the water. They're using the water. It's not like the water disappears.

So all I'm saying is the water concerns me a lot less because you can keep it inside of the system. The electricity, obviously, you use it and burn it and it's gone. We've got to figure out how to make more electrical compute. So that's me trying to be as middle of the road on that as possible.

In any case, we know there's a ton of competition. So whether Meta does this or not, people are going to criticize them for those types of issues. Whether they do it or not, other players are going to do it. So they're obviously going to compete.

We have the AI hyperscale at CoreWeave is planning a data center expansion in Dallas as well. We have, you know, the huge Stargate project, those announced by OpenAI and Oracle. We have the massive data center by XAI, the Colossus supercomputer. So all of this is going down right now.

It's going to be interesting. There was a column recently featured in The Economist on Monday and the US Secretary of Energy, which is Chris Wright, called for the US to, quote, lead the next major energy intensive frontier artificial intelligence. He said that AI transforms electricity into, quote, most valuable output imaginable intelligence. And he also was, you know, essentially talking about the fact that the federal government is going to accelerate the production of high energy derived from coal, nuclear, geothermal, and natural gas.

So evidently people in the government are focused. They can see what is happening. The writing's on the wall. Everyone's scaling and scrambling as fast as they can to build these huge data centers.

We need energy to power it. And so I think we're going to get there. But China is also doing this to quite a high degree. So we have competition and it's going to be interesting to see what happens with all of that.

So in any case, I will keep you up to date as new companies announce these data centers as they get closer to completion, as we're grappling with the energy side of it, the crisis that may arise there. All of that. Thank you so much for tuning into the podcast. If you want to try all of the latest AI models, make sure to go check out AIbox.ai.

Thanks so much for tuning in. Leave a review if you enjoyed the episode. And as always, I hope you have a fantastic rest of your day.

Frequently asked questions

How long is this episode of Hard Fork AI?

This episode is 12 minutes long.

When was this Hard Fork AI episode published?

This episode was published on July 20, 2025.

Is there a transcript available for this episode?

Yes, a full transcript is available for this episode. You can read the complete transcript on the episode page.

Can I download this Hard Fork AI episode?

Yes. Use the download control on the episode player to save the publisher-provided media file.