DDN picks up AMPHIX edge AI Factory-in-a-Box win

AI & Compute

DDN Lands HyperPOD Win in AMPHIX Edge AI Factory on Tribal Lands

DDN's HyperPOD now runs inside AMPHIX, a microgrid-powered edge AI factory on Northern California tribal lands, with 98.8 percent GPU utilization in a 24-hour RAVEL test — and no customers yet.

By
Sophie Lindqvist
Filed
Channel
AI & Compute
Read
4 min read

DDN has landed an edge AI HyperPOD deployment inside AMPHIX, a containerized AI Factory-in-a-Box built on the tribal lands of the Cachil Dehe Band of Wintun Indians in Colusa County, Northern California. The site sits west of Yuba City in the northern central valley. It is the second such factory in a planned rollout. And, as far as anyone has disclosed, it has no end-user customer yet.

The deal is a multi-player mini-Neocloud arrangement with four named technology participants and one conspicuous omission. The storage supplier for the first site, in Ironton, Ohio, was never revealed. This time DDN is on the record.

Who builds what?

Strata Expanse, a Real Estate Investment Trust founded in 2023 and headquartered in The Woodlands, Texas, develops energy-first, modular infrastructure for AI and high-performance computing. It owns a portfolio of sites with access to power resources, and there could be up to 100 sites in its network. Strata Expanse provides the land, power, cooling, connectivity and supporting infrastructure that form the physical foundation of AMPHIX.

The company operates on a "Grey Space as a Service" model. It owns, operates and maintains the non-IT infrastructure through long-term subscription agreements, which it says removes capital and execution risk for edge compute operators. Supermicro supplies the IT hardware: liquid-cooled and air-cooled clusters plus GPU/CPU platforms for large language models, GenAI, training workloads and HPC. RAVEL contributes its Orchestrate AI control plane software.

The first Strata Expanse AI-Factory-in-a-Box, branded an AI Center of Excellence, was built in December 2025 in Ironton, Ohio. It formed part of a planned 30-site national rollout of modular deployments. The storage supplier for that site remains undisclosed.

What changes with AMPHIX?

The California site departs from the Ohio template in one key respect: power. A microgrid and energy infrastructure partnership links Colusa Indian Energy, a 100 percent tribally owned clean power infrastructure company, with Strata Expanse. Plans call for expanded onsite generation — solar, battery storage and additional generation — to exceed 100 MW within 18 months.

The RAVEL AMPHIX AI Infrastructure Platform is a modular deployment model designed to bypass traditional utility grid bottlenecks. It pairs site-ready land and turnkey microgrids from Strata Expanse with certified compute hardware from Supermicro and orchestration software from RAVEL.

DDN's role centers on its Enterprise AI HyperPOD, based on Infinia object storage, with the physical kit built and supplied by Supermicro. HyperPOD is a pre-validated, pre-integrated modular AI Factory built on the Nvidia Data Platform architecture. Customers can order it in predefined sizes and deploy it rapidly.

What did the stress test show?

RAVEL has published benchmark figures for the integrated stack. In a 24-hour training test, HyperPOD sustained 98.8 percent GPU utilization with zero thermal throttling and zero GPU errors, holding 37°C of thermal headroom at peak load. RAVEL says these results give partners and customers performance evidence before making infrastructure decisions.

RAVEL Chief Product Officer Philippa Carrol said: "HyperPOD stood out to us because it represents what we believe AI infrastructure needs to become: a packaged, production-ready solution built through collaboration across the ecosystem."

She added: "By bringing together DDN, NVIDIA, Supermicro and RAVEL, we can validate the complete infrastructure stack against real-world workloads, not just individual components. At AMPHIX, we've built and tested this integrated environment and are now ready to make it available for customer POCs, giving organizations the opportunity to test their workloads, validate their configuration and build confidence before moving into production."

RAVEL says it stress-tested HyperPOD across both inference and training workloads before opening the facility to proofs of concept.

What is still missing?

No customer has been announced for either AMPHIX site. The performance figures come from RAVEL's own testing, not an independent party. The 100 MW generation target and the 30-site rollout are plans, not built capacity. DDN discloses no contract value and no deployment scale in GPU counts or rack units.

The Ohio site went live in December 2025. AMPHIX is now ready for customer POCs. Whether that pipeline converts into paying workloads will determine whether this model scales beyond two demonstration facilities.

Original: strataexpanse.com

Share this article:

More from Sophie Lindqvist

Sophie Lindqvist

Show full bio

News editor covering money, funding and market shifts around Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,125 articles

Related articles

« Previous article