220,000 Nvidia GB300 GPUs in a smaller building: SpaceXAI reveals new Minihard AI cluster

SpaceXAI continues to expand its computing presence in Tennessee, unveiling a new data center called Minihard that will be equipped with 220,000 Nvidia GB300 accelerators — the same number of GPUs planned for the larger Macrohard facility.

The key difference with Minihard is its compact design. Thanks to denser server equipment layout, the company will fit comparable computing resources into a smaller footprint, reducing the building's size and improving overall infrastructure efficiency.

To ensure stable power supply, the facility will rely on Tesla Megapack battery systems. These units are designed to smooth load fluctuations and provide backup power during operation of large-scale computing clusters. SpaceXAI has previously incorporated Megapacks into its AI infrastructure projects as well.

The construction is part of the company's broader push to grow its compute capacity. SpaceXAI is developing multiple AI clusters across Tennessee, including Colossus 2 near Memphis — currently described as the world's largest supercomputer — which is expected to host up to 1 million GPUs to accelerate AI model training and inference.

The first phase, Macrohard, will feature over 330,000 GPUs, a combined performance of 4.4 trillion operations per second, power consumption exceeding 1 GW, and 558 Tesla Megapacks with enough combined capacity to support such heavy workloads. The Minihard facility is expected to mirror this capacity in a significantly more space-efficient form.