xAI reportedly targets 10 GW data center capacity by 2027 with $500 billion revenue goal

xAI’s announced 7x increase in data center capacity to 10 GW by 2027 represents a step-change in AI infrastructure. Current clusters, such as those used by xAI today, draw 1.4 GW but deliver far less usable compute due to power distribution losses, cooling overhead, and non-accelerator components. The planned expansion would require deploying millions of Nvidia Rubin GPUs, assuming a PUE of 1.2 and 70 to 80% of IT power allocated to accelerators. This scale would dwarf existing supercomputers, with projected FP64 performance 3.7 to 4.3x higher than the entire Top 500 list combined.

The technical and logistical challenges of this expansion are substantial. Cooling, power delivery, and grid integration for 10 GW of IT load are untested at this scale, particularly for AI workloads with high power density. xAI’s revenue target of $300 to 500 billion implies a per-watt value of $30 to 50, which assumes either premium pricing for AI compute or a dramatic increase in model efficiency. Neither is guaranteed, and the economics depend on sustained demand for large-scale AI training and inference, which may face diminishing returns or regulatory constraints.

The broader implications for the AI hardware ecosystem are significant. xAI’s reliance on Nvidia’s Rubin architecture locks it into a single vendor’s roadmap, with potential supply chain risks. The expansion also intensifies competition for data center sites with sufficient power and cooling infrastructure, potentially driving up costs for other operators. If successful, xAI’s cluster could enable AI models far beyond current capabilities, but the feasibility of monetizing such compute at the projected scale remains speculative.

Energy and sustainability concerns are unavoidable at this scale. A 10 GW data center would require dedicated power plants or grid upgrades, with cooling systems consuming hundreds of megawatts. The environmental impact of such a facility could attract regulatory scrutiny, particularly in regions with carbon emissions targets. Additionally, the projected performance gains assume near-perfect utilization, which is difficult to achieve in practice due to workload variability and hardware failures. The gap between nameplate power and realized compute underscores the uncertainty in xAI’s revenue projections.