Mark Zuckerberg is in full founder mode again. This time it’s the AI version.
Meta fell behind in the generative AI race earlier this year when the company rolled out its Llama 4 model. The product was not up to scratch, compared to rival offerings from DeepSeek, OpenAI, Anthropic, and Google.
Since then, Zuck has gone on a hiring bender, personally recruiting top AI talent with pay packages in the $100 million range and more.
That’s just one ingredient for success in generative AI. Another is high-quality data. Hence, Meta’s weird $14 billion deal to buy just under half of Scale AI and its leader, Alex Wang.
The third ingredient required is infrastructure, which is tech jargon for the AI chips (GPUs, etc), networking gear, and data centers needed to build, refine, and run giant AI models such as the Llama series.
On Friday, SemiAnalysis blew the lid off Zuck’s big plans for Meta’s huge new AI infrastructure. You can read the report here. You’ll need to pay to read the whole thing, but if you’re serious about AI, you will prob want to shell out.
Zuck confirmed some of the details of SemiAnalysis’s report via a Facebook post on Monday. The CEO said Meta will build several new AI data centers that use more than 1 gigawatt of power each. This is how data center capacity is measured, and anything in the 1 gigawatt range is absolutely massive (or at least it was until now).
Related stories
Business Insider tells the innovative stories you want to know
Business Insider tells the innovative stories you want to know
Data centers in tents
The thing that really caught my eye in the SemiAnalysis report is that Meta is building some of these AI data centers in tents right now. A Meta spokesperson confirmed this, and noted that the tents help the company add additional capacity to its data center campuses and stressed that the entire “supercluster” data center facility isn’t just a tent.
That’s good, because data centers house a lot of complex, expensive gear and this gear needs to be kept cool in really controlled ways, otherwise it might will overheat.
So, building parts of data centers in tents is a sign of how quickly Zuck is moving to get additional AI compute capacity up and running.
Remember when Elon Musk had Tesla making Model 3 cars in tents outside the company’s Fremont factory in 2018? That was about speed to market, and Zuck is now doing this for data centers — likely taking a page from Elon’s strategy book.
“Inspired by xAI’s unprecedented time-to-market, Meta is embracing a datacenter design that prioritizes speed above all else,” Semianalysis wrote in their Friday report. “They’re already building more of them! Traditional datacenter and real estate investors, still somewhat reeling from xAI’s Memphis site and time to market, will be shocked yet again.”
“From prefabricated power and cooling modules to ultra-light structures, speed is of the essence,” it added.
Tents get really hot, so this could be a challenge for running the prefab parts of these AI data centers. Indeed, Semianalysis reported that Meta could shut down workloads during the hottest summer days.
Over the long term, Meta will likely build full data centers without tents, but in the short and medium term, the company needs these facilities up and running “as soon as possible,” Dylan Patel, CEO of SemiAnalysis, told me on Monday.
“Everyone is trying to build data centers as fast as possible in the race to achieve AGI. Due to constraints with power, datacenter capacity, and the construction crews, Meta has started putting datacenters in ‘tents’ to reduce construction bottlenecks,” Patel said.
And here’s the full post from Zuck on Monday:
“For our superintelligence effort, I’m focused on building the most elite and talent-dense team in the industry. We’re also going to invest hundreds of billions of dollars into compute to build superintelligence. We have the capital from our business to do this. SemiAnalysis just reported that Meta is on track to be the first lab to bring a 1GW+ supercluster online.
“We’re actually building several multi-GW clusters. We’re calling the first one Prometheus, and it’s coming online in ’26. We’re also building Hyperion, which will be able to scale up to 5GW over several years. We’re building multiple more titan clusters as well. Just one of these covers a significant part of the footprint of Manhattan.
“Meta Superintelligence Labs will have industry-leading levels of compute and by far the greatest compute per researcher. I’m looking forward to working with the top researchers to advance the frontier!”
Sign up for BI’s Tech Memo newsletter here. Reach out to me via email at abarr@busienssinsider.com.
