RT-One Certifies AI Stack Once to Unlock Multi-Country Campuses

RT-One has signed on as the first commercial customer of the AMPHIX AI Infrastructure Platform from RAVEL and Strata Expanse, using a Centre of Excellence near Colusa, California, to certify one production stack it will copy across planned AI campuses in the Americas.

The move lets the company stand up customer environments and begin serving users from the live COE while permanent campuses remain under construction, compressing the gap between capital outlay and revenue.

How the AMPHIX Deal Works for RT-One

RAVEL and Strata Expanse announced the agreement on August 11, 2026. AMPHIX integrates prepared land, resilient power, cybersecure edge networking, certified compute stacks and orchestration software into a single deployment model.

RT-One develops federated AI infrastructure campuses and already partners with RAVEL on a sovereign cloud approach for the region. Work at the Colusa COE will lock down the reference configuration for that model. Once proven, the compute, network, thermal and orchestration profile transfers to each new site, with changes limited to local regulations.

• Strata Expanse supplies the physical base: land, power, cooling and secure connectivity under its Gray Space as a Service model.
• Available Infrastructure adds national zero-trust edge networking.
• RAVEL certifies the technology ecosystem, validates designs, trains operators and runs Orchestrate AI for workload management and policy.
• Certified partners include Supermicro, DDN, NVIDIA platforms and others for the compute and storage layers.

Denise Muyco, CEO of RAVEL, said the complexity of AI stacks and GPU demand have raised capital risk and operational exposure. AMPHIX Centres of Excellence let customers validate against real workloads before major outlays.

Traditional Pilots Cost Millions and Still Miss Production

Conventional production-grade pilots can take up to nine months and cost as much as $4 million, yet often fail to test the exact stack that ends up in production. Organizations then scale on incomplete information.

AMPHIX flips the sequence. Customers validate the precise environment they will run, tune models and policies, train teams, and in some cases start delivering service from inside the COE. RAVEL and Strata position the approach as a modular validated path from land to production AI.

Colusa Tribal Microgrid Supplies the Power Layer

The Colusa COE ranks among the largest planned AMPHIX sites. Power comes from a microgrid developed and managed by Colusa Indian Energy, owned by the Cachil Dehe Band of Wintun Indians of the Colusa Indian Community.

Colusa Indian Energy and Strata Expanse announced their partnership in March 2026 as Strata’s first with a Tribal Nation. Proof-of-concept work interconnects immediately to the community’s existing microgrid. Plans call to expand onsite generation past 100 MW in 18 months to support the COE.

• 23+ years operating the flagship microgrid in island mode off-grid.
• 14 years without a single power outage on that system.
• Team experience delivering more than 2.5 GW across 100-plus countries.
• Immediate interconnection for partner POCs while capacity scales.

That reliability track record matters. Grid interconnection queues for large AI loads now stretch years in many U.S. markets, which is why more AI operators turning to on-site power solutions are looking at microgrids and dedicated generation. The tribal land setting also carries sovereignty and permitting characteristics distinct from ordinary commercial sites.

Greg Wood, senior director at Strata Expanse, said AMPHIX gives customers a more repeatable blueprint. Strata handles the physical foundation so operators can focus on compute environments.

RT-One’s Multi-Country Campus Bet

Fernando Palamone, CEO of RT-One, framed the commercial pressure clearly. Customers buy contracted capacity with firm delivery dates and defined performance standards. The company pays a price for missing either.

RT-One has already secured land and energy strategies across the Americas. One flagship project is the Uberlândia campus in Minas Gerais, Brazil: more than $1.2 billion planned investment on over one million square metres, phase one at 100 MW scaling toward 400 MW total, designed for 100 percent renewable power and positioned as Latin America’s largest AI data centre. Construction partners have included Hitachi Energy and others.

Additional sites are referenced for Paraguay, Colombia and further markets under the federated sovereign model. The COE work certifies the standardized deployment block that will travel with only regulatory adaptations.

1. March 2026, Colusa Indian Energy and Strata Expanse partner; Amphix COE planned on tribal land with microgrid expansion.
2. March 2026, RAVEL, Strata and Available formally introduce Amphix platform ahead of NVIDIA GTC 2026; Ohio pilot COE referenced.
3. June 2026, RT-One and RAVEL deepen partnership on federated sovereign AI infrastructure across the Americas.
4. August 11, 2026, RT-One named first commercial AMPHIX customer; Colusa COE becomes the certification vehicle.

Every validation hour completed in California is an hour not repeated in Brazil or elsewhere. That is the second-order compression: not faster construction on any single site, but one certified profile serving many.

One Stack, Many Jurisdictions

We have secured land and the energy strategy across the Americas. What the COE gives us is the ability to certify the exact configuration we intend to deploy – the same standardized block, the same stack, the same operating policies – while the campuses are under construction, rather than waiting until after they come online. Every validation hour we complete in the COE is an hour we do not spend proving the same point again in Brazil, in Paraguay, or in Colombia. That is what compresses the interval between capital commitment and revenue service: not speed on any single site, but certifying once and deploying many times.

Fernando Palamone, CEO of RT-One, said the above in the joint announcement. The same environment supports model tuning, operational policy setting for performance, energy and cost, and training of ops teams on the final production stack.

RAVEL’s Orchestrate AI handles workload management across the validated setup. Because the COE runs the identical orchestration and governance frameworks planned for production, validated workloads move without full redesign.

Why AI Infrastructure Operators Face This Pressure Now

GPU demand, liquid cooling requirements and power densities have turned infrastructure into the binding constraint. Industry analyses put the multi-year AI data-centre build-out in the multi-trillion-dollar range. One detailed breakdown shows that for a typical one-gigawatt AI facility, servers dominate one-gigawatt AI data center TCO at roughly 60 percent of annualized ownership cost, with energy and other OpEx far smaller.

Delays carry direct opportunity cost. Operators selling firm capacity cannot slip delivery dates without penalties or lost contracts. Traditional build-then-test sequences leave that risk open until late. AMPHIX’s production-grade Centers of Excellence for AI validation aim to close it earlier by matching pilot density, power and thermal conditions to production.

The model also targets neo-clouds, regulated industries needing sovereign control, and enterprises moving from proof-of-concept clusters to full AI factories. Workforce enablement sits inside the COE mandate so operators arrive trained.

Early Revenue While the Cranes Still Turn

For RT-One the immediate objective is clear: customer environments online months ahead of campus completion, faster onboarding, lower deployment risk, and a single certified block that travels. The Colusa microgrid supplies the reliable, expandable power foundation that makes the live validation environment credible.

If the certification holds, the company can start metering usage from California while concrete pours continue in Uberlândia and later sites. That changes the cash-flow profile of multi-country AI campus development more than any single construction acceleration ever could.

RAVEL and Strata Expanse now have their first commercial reference. The rest of the market will watch whether one proven stack really does deploy many times without expensive re-learning.