Elon Musk says xAI bought a third building called “MACROHARDRR”, reportedly adjacent to Colossus 2, that will take the company's training compute to almost 2GW
Elon Musk's xAI is planning an expansion of its massive data center complex in Memphis, and has purchased a third building …
Context & Ripple Effects
xAI’s latest building purchase extends a rapid Memphis buildout that began with a 100,000-H100 Colossus cluster and later moved toward a much larger Colossus expansion. The company had already been reported to be pursuing roughly 300,000 additional Nvidia chips for Colossus 2.
The planned MACROHARDRR site makes the buildout a multi-facility regional footprint rather than a single-supercomputer project; subsequent coverage describes it as xAI’s third data center in the greater Memphis area.
First-order effects
- xAI gains another site for training infrastructure, with Musk saying the combined training-compute footprint will approach 2GW.
- The expansion increases near-term demand for the hardware, electrical equipment, and grid-support systems needed to turn additional building capacity into usable AI compute; xAI had already used Tesla Megapacks to support Colossus.
Second-order effects
- Nvidia and the surrounding data-center supply chain gain another concrete outlet for xAI’s planned accelerator procurement, following the reported Colossus 2 chip expansion.
- A larger clustered load raises the importance of power availability and on-site resilience in the Memphis-area buildout, potentially making energy infrastructure a gating factor alongside accelerator deliveries.
Third-order effects
- If comparable expansions continue, frontier-model competition will increasingly be shaped by firms’ ability to secure long-lived power, sites, and financing—not solely by model development.
- The project is another instance of AI training infrastructure being built as utility-scale industrial capacity, concentrating demand and operational dependencies in a small number of large regional campuses.
The trend: Frontier AI developers are moving from discrete GPU-cluster deployments toward regional, power-constrained compute campuses built around durable infrastructure commitments.