Xeal wants to turn idle EV chargers into AI inference hubs

Introduction

Xeal, a New York-based charging operator, has launched Laitent, the world's first edge inference compute network built on spare electrical capacity at EV charging stations. The company plans to deploy more than 100,000 Nvidia GPUs across its charging network, targeting response times under 20 milliseconds in metropolitan areas.

What Xeal is Building

Xeal runs one of the five largest Level 2 charging networks in the US. Its footprint covers more than 1,600 chargers spread across more than 500 cities. Behind those plugs sits over 200 MW of permitted, grid-connected capacity. During charging hours, that capacity typically runs at less than 10% utilization. Laitent is meant to fill that empty lane. The vehicle is a compact unit called a Laitent Pod, sized to fit into a single parking spot. Each Pod can hold up to 48 Nvidia GPUs from the Hopper or Blackwell Ultra lines. The company says the Pods are designed to sit outdoors, need no water connection, and stay below 65 dB in noise.

Partnerships and Funding

Xeal has lined up partners to help. The company has announced collaborations with Rafay Systems and Spectrum Business. Its first Laitent Pod installation is targeted for the end of 2026, alongside JVM Realty. Xeal was founded in 2019. Since then it has raised approximately $54M to $64M in venture funding, including a $40M Series B round in 2024. Laitent's ambitions sit on a much larger scale. Xeal aims to raise around $2 billion in project financing over the next 18 months.

Benefits for Property Owners

For property owners and real estate partners like JVM Realty, the pitch is straightforward. Host a Pod, keep the EV chargers running, and potentially see a boost in property value of up to $1 million per installation, per Xeal's own estimate. Xeal needs to close roughly $2 billion in project financing, a figure far beyond its venture history. It needs its orchestration software to genuinely keep drivers first. And it needs enough paying inference customers to justify a fleet that could eventually exceed 100,000 GPUs.

Milestones to Watch

The milestone to watch is the first JVM Realty installation, targeted for the end of 2026.