EV charging company Xeal plans to deploy 100,000 Nvidia GPUs in pods at roadside charging sites across the U.S., creating a distributed edge inference network.

EV charging firm Xeal plans to leverage its existing U.S. network to deploy what it describes as the "world's first edge inference compute network using idle EV charging capacity." The company operates over 1,600 locations across the country with 200MW of permitted, installed electrical infrastructure. Its strategy centers on deploying more than 100,000 NVIDIA GPUs through custom enclosures called Latient Pods, each capable of housing up to 48 NVIDIA Hopper or Blackwell Ultra GPUs.

Nikhil Bharadwaj, Xeal's co-founder and CEO, framed the initiative as addressing a critical bottleneck in compute expansion. "Latient is the fastest way to bring new compute online, tapping already permitted and grid-connected, yet idle energy capacity, located right where it's needed for ultra-low latency, with no water hookup requirements," Bharadwaj said. The company's EV charging sites, which typically operate at less than 10% of permitted capacity, can deliver processing power through a distributed network spanning metropolitan areas—sidestepping grid interconnection, one of the most challenging obstacles for new data center development.

NVIDIA has endorsed the approach, with a company representative praising Latient Pods as an innovative method to bring low-latency inference closer to users while making fuller use of existing infrastructure.

The Latient Pods themselves are designed for durability and simplicity. Rendered in images as resembling broadband fiber cabinets, each pod features self-contained independent cooling, requires no water hookup, and operates at under 65 decibels—comparable to a washing machine. Units arrive in a NEMA 4 outdoor-rated enclosure, can be installed and running in hours, and require no water hookup. Power costs are covered as a tenant pass-through, and site owners receive fixed rental income or ancillary revenue, with potential property value appreciation of up to $1 million per pod.

With 1,600-plus existing locations and each pod containing up to 48 GPUs, Xeal's target of over 100,000 GPUs implies deployment of multiple pods at some sites or expansion of its site footprint to around 2,100 locations. The pods' positioning across metro areas—a feature Xeal calls "Metro Edge"—enables sub-20-millisecond latency, a substantial improvement over inference hosted in distant data centers.

The distributed GPU network echoes earlier proposals in the EV sector. Last year, Elon Musk proposed tapping Tesla's fleet of millions of vehicles as an untapped computing asset, calling it "100 gigawatts of inference" available when vehicles are idle. Xeal's approach channels that concept through existing charging infrastructure instead.

The security implications, however, warrant attention. EV charging stations have already become theft targets, with cable theft driven by copper scrap value. A Latient Pod containing 48 Blackwell GPUs would represent approximately $2 million in hardware value—a substantial asset requiring robust protection to deter sophisticated theft.

Xeal plans to have the first Latient Pod online by the end of the year.