Elon Musk told SpaceX employees that xAI will expand data center power capacity sevenfold to 10 gigawatts by the end of 2027, from current capacity of 1.4 GW. Musk projected that reaching 10 GW with a value-per-watt price of $30–$50 could generate $300–$500 billion in annual revenue by the end of next year. The company plans to deploy between 2.1 and 2.4 million Nvidia Rubin GPUs in Max-P mode, with expected inference performance reaching 105–120 exaFLOPS—roughly 3.7–4.3 times the combined FP64 performance of all Top 500 supercomputers combined. This buildout positions xAI's cluster among the largest AI training and inference infrastructures globally.
The 10 GW target represents an 8.6 GW increase in nameplate power draw added to existing infrastructure across Memphis and Southaven data centers. Given typical facility overhead (power usage effectiveness of 1.2) and allocations for CPU, memory, and networking, actual accelerator capacity would likely receive 4.8–5.5 GW of the new power budget. Tom's Hardware's analysis suggests xAI is banking on Nvidia's Vera Rubin systems, with NVL72 VR200 rack-scale systems consuming approximately 2.3 kW per GPU when configured at maximum performance.
Architects tracking AI infrastructure capacity need to monitor whether xAI can secure the electrical supply, cooling, and physical space to meet this timeline. xAI's revenue model depends on both sustained demand for frontier compute and competitive pricing—with NVIDIA, CoreWeave, and Lambda already establishing baseline rates. The $300–500B revenue claim assumes 100% utilization and premium pricing; real-world execution will be crucial to validate whether the company can both deliver the hardware and maintain pricing power against rivals ramping their own megafabs.