Distributed AI cloud provider Perimeter Compute launches to deploy GPUs at the Edge

A new AI cloud provider has entered the market with a distinctive approach to edge computing. Perimeter Compute, a portfolio business of Montauk Capital, announced its launch this week with plans to deploy GPU infrastructure inside underutilized commercial office buildings across major U.S. cities, tapping into excess energy capacity that would otherwise go unused.

The company is led by David Hall, a former executive at GPU cloud provider Lambda. Hall told Latitude Media that Perimeter Compute intends to install GPUs in the basements and mechanical rooms of office buildings, targeting sites with between 500kW and 20MW of spare power capacity. He noted that a single 500kW allocation alone can power 240 of the latest Nvidia GPUs, underscoring the scale of compute that can be unlocked from existing building infrastructure.

In a LinkedIn post on August 3, Hall framed the launch as a response to surging demand for AI inference and the mounting challenges facing traditional data center development. "Today, we're launching Perimeter Compute to the world to deliver the shortest path to data center-capable GPUs at the Edge. With AI inference demand skyrocketing and large-footprint data centers facing regulatory and supply chain hurdles, we provide a faster alternative: we deploy within existing, powered facilities to get companies inference compute in weeks, not years," he wrote.

Hall emphasized that the company is not simply another distributed compute player. "Plenty of players are chasing distributed and Edge compute. We are doing something different: delivering a technical and financial solution from day one that fits within existing and underutilized space without the need for costly capex. We partner directly with real estate owners and operators, creating immediate value for them and our customers. And, our inference hubs are equipped to provide enough capacity to run even the largest frontier-class models at the Edge," he added.

The launch comes at a time of growing momentum in edge AI deployments. Several companies have begun rolling out Nvidia's AI Grid reference design to build distributed networks of GPU compute, with telecommunications providers leveraging their existing infrastructure footprints to host edge capacity. Akamai has also adopted the approach, utilizing its network of 4,400 points of presence to support distributed AI workloads.

Perimeter Compute's model represents a broader shift in how AI infrastructure is being sourced and deployed. By repurposing existing commercial real estate with available power, the company aims to bypass the lengthy permitting and construction cycles that have slowed traditional data center projects. The approach also offers building owners a new revenue stream from idle capacity, while giving AI companies a faster, lower-capital path to inference compute at the edge. As demand for low-latency AI inference continues to grow, the success of this model could signal a meaningful evolution in how distributed compute resources are brought online.