Xeal proposes an AI inference network powered by the electrical capacity left unused at its US EV charging locations. Its target exceeds 100,000 NVIDIA GPUs. Xeal expects its first cabinet to begin operating before 2026 ends. Existing grid connections underpin the plan, but cabinet distribution and the route to the full GPU target remain unspecified.
Artificial Intelligence··Morning
Spare charging-site power would run inference cabinets
EV charging company Xeal plans to place AI inference cabinets at existing sites across the US, using spare electrical capacity for running models. Xeal puts its permitted and installed electrical infrastructure at 200 MW, spread across more than 1,600 locations. It says charging activity typically uses less than 10 per cent of permitted capacity. Its deployment target exceeds 100,000 NVIDIA graphics processors, and Xeal expects to activate its first cabinet before 2026 ends. One Tom’s Hardware report carries this deployment plan.[1]
Each pod is designed for up to 48 GPUs
The Latient Pod cabinet would hold NVIDIA Hopper or Blackwell Ultra processors, with a maximum of 48 GPUs. It uses self-contained cooling without an external water connection and an outdoor enclosure. Xeal says installation can take hours. Its Metro Edge arrangement would place computing near metropolitan users, with a company-claimed latency below 20 milliseconds.[1]
The full GPU target requires more than one cabinet per site
A single cabinet containing 48 NVIDIA GPUs at each of 1,600 sites would fall short of the stated target. Xeal would need multiple cabinets at some sites or a larger site network; the distribution has not been disclosed. Site owners would receive fixed rent or ancillary revenue, while electricity bills are intended to pass through to computing tenants.[1]
Related columns
For more information on this topic, you can read the related columns.
