Available’s $5B Project Qestrel aims to roll out 1,000 AI-ready edge data centers by year’s end

Hyperscaler data center projects are plagued by a host of issues: delayed time-to-market, capacity constraints, and supply chain issues.

However, secure edge infrastructure provider Available Infrastructure aims to offer a different experience: A “nationwide fleet of cybersecure, private neocloud edge data centers.” Through a new $5 billion initiative, Project Qestrel, the company has an ambitious plan to bring 1,000 locations in 100 US cities and 30-plus states online by the end of this year.

Each site will be equipped with high performance compute (HPC) infrastructure and AI inferencing capabilities, which the company says will bring sites online “in weeks to months, not years.”

Project Qestrel is built around three fundamentals, according to Available’s EVP of strategy Dan Medina: speed (ultra-low latency), location (close to the mission), and security (zero trust with quantum-resilient encryption built-in).

“We view this as the beginning of a shift in AI infrastructure: Inference moving to the edge, under a security model built for modern threats and a future shaped by post-quantum risk,” he said.

Modularizes deployment

What’s unique about Available’s offering is that each deployment will be co-located at a telecom site, providing “immediate access” to power and fiber. This eliminates the “longest critical-path items,” such as land acquisition, substation interconnect queues, and multi-year power delivery timelines, Medina noted.

Available is partnering with wireless infrastructure company Crown Castle, which owns, operates, and leases more than 40,000 cell towers and roughly 90,000 miles of fiber.

“Our strategy is to industrialize and modularize deployment by building on telecom co-location and pre-existing physical infrastructure rather than greenfield hyperscale construction,” said Medina.

Some initial sites are live (the company declined to say how many, due to “final contractual and commissioning milestones”) and 30 cities are expected to come online by early July. Available is prioritizing dense urban corridors, and early adoption has begun in “major Northeast corridors with a path to nationwide rollout,” Medina explained.

The company’s infrastructure will be used by Strata Expanse, which specializes in 60 to 90 day AI data center deployments, and incorporated into Strata’s new full-stack, end-to-end Amphix AI Infrastructure Platform.

The neocloud architecture will run up to 48 GPUs per site, bringing AI inferencing to the edge. Many sites will be pre-integrated with IBM’s watsonx; others will be AI-agnostic, allowing enterprises to run their preferred models.

According to Available, Project Qestrel will provide:

• Edge-first architectures that keep data where it’s created to protect sensitive information and meet residency requirements;
• Zero-trust and quantum-resilient encryption to protect against emerging threats;
• HPC at the edge for “ultra-low-latency” applications;
• Air-cooled, power-ready sites to support fast deployment.

Yaz Palanichamy, a senior advisory analyst at Info-Tech Research Group, pointed to the “sheer size” of this deployment. “Having 1,000 individual physical sites across 100 major US cities is a monumental undertaking,” he said. This is especially true considering the capex outlay and the combined resourcing and people power required to plan large-scale data center deployments.

“An undertaking of this size highlights the increasing demand for distributed infrastructure to support ever-growing computing workloads from artificial intelligence and low-latency applications,” he said.

Avoid the ‘round-trip’ problem

The fastest-growing AI workloads for many industries are inference and real-time decisioning, Medina noted, but those can fail when they have to “round trip” to distant hyperscale regions. Telecom co-location aligns to a “practical reality”: Many enterprises can’t wait for multi-year mega-campus timelines.

“In today’s geopolitical and supply-chain environment, speed is not optional,” said Medina. “The ability to bring capacity online in weeks to months, not years, is a core design requirement and a market differentiator.”

Available’s SanQtum technology is designed as a “zero trust mesh,” Medina explained. Every user, device, and workload is continuously verified, and network access is micro-segmented by identity and policy, not by location. The platform integrates with enterprise identity providers such as Entra and Okta to support continuous monitoring, policy enforcement, and audit-ready logs.

Further, said Medina, “shared telecom locations are handled via strict tenant isolation, identity-bound policy enforcement, and auditable controls, so customers can share physical proximity without sharing trust boundaries.”

A model that ‘extends naturally’

Project Qestrel is structured as a phase-based national buildout intended to be “future-proof” beyond phase 1. Medina noted that the growth model “extends naturally” because the platform is modular and repeatable.

This means that “adding sites is an exercise in provisioning and integration, not redesign,” he said. The company can also expand tower and telecom footprints where lease options exist, supporting additional phases as demand and customer commitments scale.

Lastly, Medina noted, “the architecture is quantum-forward, given its infrastructure and the fact that it standardizes secure connectivity, identity, and post-quantum cryptography at the edge.”