Each Latient Pod holds up to 48 Nvidia Hopper or Blackwell Ultra GPUs in a self-contained, weatherproof enclosure with its own cooling and no water hookup. Xeal says the pods run quietly at under 65 dB, comparable to a washing machine, can be installed and running in hours, and offer site owners fixed rental income, up to $1 million in added property value and pass-through power bills. The pitch is speed: grid interconnection is one of the hardest parts of standing up new compute, and Xeal's 1,600-plus locations, with 200MW of permitted electrical capacity, typically run at less than 10% of what they are allowed to draw. Chief executive and co-founder Nikhil Bharadwaj says that makes Latient the fastest way to bring new compute online, delivering inference now while purpose-built data centres are still under construction.

The company claims its pods scattered around metro areas, a feature it calls Metro Edge, can offer sub-20ms latency, well ahead of inference run from a distant data centre. An Nvidia representative praised the plan as an innovative way to move low-latency inference closer to users while making use of infrastructure that already exists. The arithmetic is loose, though: 1,600-plus sites multiplied by up to 48 GPUs per pod falls short of 100,000, so either some locations will host multiple pods or Xeal expects to build out to roughly 2,100 sites.

There is a practical wrinkle: EV charging sites are already targets for thieves, with Tom's Hardware previously reporting a wave of charger cable thefts driven by the scrap value of copper, and a pod holding up to 48 Blackwell GPUs would be worth roughly $2 million or more at 2024 prices. Xeal says the first Latient Pod will be online by the end of the year.