xAI Runs the World's Most Powerful Compute Cluster 🤖

Key Takeaways

• With an estimated 200,000 Nvidia H100 chips, xAI has the largest compute cluster globally located in Memphis, TN.
• xAI’s compute cluster can hit 20.6 on a log₁₀ scale for non-sparse operations per second, which is over 400 quintillion calculations each second.
• Meanwhile, Meta, Oracle, and Microsoft/OpenAI run compute clusters with 100,000 H100 equivalents.

Supercomputers Are Getting More Powerful and Expensive

Today, xAI's Colossus supercomputer uses an estimated $7 billion in hardware—outpacing Meta's 100K by around $3 billion.

Overall, acquisition costs of the most powerful supercomputers grow by twofold every 13 months.

Following a similar trend, the power demand for these compute clusters has increased by 1.9 times each year since 2019, with some local grids unable to provide enough power capacity to a single location.