Meta Is Building the World’s Largest AI Cluster, Which Will Come with Compute Power Never Seen in the Markets; Set to Feature a “Tent-Like” Structure for Rapid Deployment

Muhammad Zuhair
Person in sunglasses in front of a large data server room, highlighting tech and computing.

Meta's next plans for the AI markets might be something no one could even imagine, as the company's next cluster would be the world's most powerful one.

Meta's First GW-Scale AI Cluster Is Expected to Come Live By 2026, Being Faster Than OpenAI's Stargate

Meta has been making massive moves, probably one of the largest ones in the AI segment, and it seems like there's no stopping them for now. Zuckerberg has managed to switch up gears with the AI hype, and now the goal is simple: invest and dominate. After building an "Avengers-level" team, Meta now plans to ramp up its hardware efforts, as the company has intentions to build a "gigawatt" AI cluster called Prometheus, and it will be something that the industry has never seen before, with the project expected to beat even xAI when it comes to deployment.

Related Story Mark Zuckerberg Reveals Meta’s Plans for “Multi-GW” AI Clusters in the Coming Years; Will Be Paired With the “Avengers-Level” AI Team

This project is a one-of-a-kind solution, offering performance higher than that of Stargate, which is valued at at least $100 billion. More importantly, Meta plans to utilize natural gas as the main source of power, building up to 200MW on-site natural gas plants. In terms of equipment, Prometheus is expected to employ up to 500K GB200/GB300 AI clusters by NVIDIA, which could be alone worth trillions when you consider the HW value, but Meta likely has leasing contracts squeezed in between, so the actual spending from the company might not be similar to the hardware value.

Prometheus is expected to be a breakthrough in the realm of AI computing, and the cluster is expected to bring in 3.17 trillion TFLOPS of power, which is higher than Stargate. But, if you want to factor in the performance relative to a modern-day supercomputer, it is nearly twice as powerful as Frontier, and interestingly, the supercomputer took almost three years to be operational, while Meta's first GW-scale cluster might take around 12-15 months. So, this shows how big of a leap has computing taken over a span of a few years.

Meta is making giant moves right now, and as the company bumps up its CapEX with each quarterly report, it is clear that they aren't looking for mediocrity. Apart from Prometheus, the firm has plans to put a "2GW-cluster" live as well, named Hyperion, and this shows that the AI bandwagon has a lot of distance to cover for now.