Get compute

The distributed cloud

Powered capacity at the speed of demand.

Nawa is building a network of modular AI compute on power that already exists, in the United States and abroad, with each cell live within a few months of signing. The pockets of generation too small or too remote for a hyperscaler add up to a footprint the market could not reach before.

Stranded power

Thousands of megawatts sit behind the meter.

The pockets are operating plants with headroom, decommissioned plants whose switchgear is still live, industrial sites with a service in place, and international capacity in markets where the demand never arrived. A 7 MW site is noise to a hyperscaler, and the interconnection queue makes a greenfield build a multi-year bet. Because this power is already generated and conditioned, there is no queue to wait in. Deployment is gated by container lead time.

  • Operating plants

    Generation with headroom the grid is not taking.

  • Decommissioned plants

    Live switchgear and a yard with nothing behind it.

  • Industrial sites

    A service in place, sized for a load that left.

  • International capacity

    Abundant power in markets where demand never arrived.

The demand that fits

The fastest-growing workloads want to be placed.

Frontier training will keep concentrating in contiguous gigawatt campuses. That is a different market. Inference and fine-tuning scale with the applications they serve, so they move toward users. A robot runs its models wherever the robot is. Sovereign buyers need capacity inside their own borders on their own power, and that demand cannot be served from a centralised campus. As the industry mix shifts from training toward deployed inference, the market moves toward this architecture.

Speed

The schedule is set by the factory.

A conventional developer waits for the interconnection. When the power is already conditioned and the offtake is already signed, adding the next megawatt is a manufacturing and logistics problem, and manufacturing problems get cheaper with volume.

  • NAWA F2 cell About 3 months

    Factory lead time for the two containers and the cooler skid.

  • Interconnection queue About 5 years

    Median time from request to commercial operation for generation projects built in 2023.

NAWA F2 cell: Nawa white paper, September 2026. Interconnection queue: Lawrence Berkeley National Laboratory, Queued Up: 2024 Edition.

NAWA F2

The cell,
part by part.

Compute, cooling and power. Built to work together.

Racks Concept
Select a part to take a closer look.

The compute inside.

Vera Rubin NVL72 racks sit inside the IT module. We coordinate the rack layout and integration with our silicon partners.

Explore NAWA F2

The cell

One megawatt in two standard containers.

NAWA F2, plan and elevation Plan: an IT container with an incoming panel, a heat-exchanger bay, six compute rack positions, three fabric positions and an air plant, beside a power container with a switchboard, a UPS and a battery behind a fire partition, and a transformer on the pad. Elevation: the IT container on its slab with a V-bank cooler on a roof frame carried by the corner castings, and the stacked option inset. service aisle to next cell 20 ft where cells never stack stacked option 05 03 01 02 04 06 07 08 09 10

Plan and elevation. F2 layout, September 2026. Not to dimension.

Hover to explore. Select a component to pull it out.

The deployable unit is a two-container cell in standard ISO boxes: an IT box carrying six NVL72-class racks with their fabric, and a power box carrying the switchboard and the UPS. The pair is tested together at the factory and re-joined on site with one busduct connection. It stacks from one cell to a campus, and the module set stays the same across geographies and voltage standards.

Open NAWA F2

Silicon partnerships

Racks shipped in as little as four months.

We work closely with silicon partners to ship racks in as little as four months, coordinating hardware supply with the infrastructure that will power and cool it. Timing depends on configuration and availability.

Discuss a deployment →

Contracted before built

Offtake is signed before construction starts.

The model contracts capacity with an established GPU-cloud partner at full utilisation over a multi-year term before the first cell is ordered. That is a rule. It turns marketplace risk into a deployment-speed problem, and deployment speed is the variable the cell is engineered to control.

Pipeline and design basis

Each new geography makes the last one worth more.

As nodes accumulate, capacity becomes a standard product: contracted in-country capacity, live within a quarter. Every new geography adds jurisdiction coverage and latency coverage to every existing customer relationship, so the platform compounds. Nawa ends up holding the sites, the power agreements, the offtake relationships and the operating record, with the option to move up the stack.

The development pipeline North-west coast, Sri Lanka: Phased deployment, In diligence. Dallas–Fort Worth, United States: Under review, In diligence. San Francisco Bay Area, United States: Under review, Under review. United Arab Emirates, United Arab Emirates: Under review, In diligence. Ankara region, Türkiye: Under review, Under review. Dodoma region, Tanzania: Under review, Under review.

Development pipeline

  • 6 Opportunities Across site diligence and regional assessment
  • 5 Countries United States and four international markets
  • Sri Lanka Flagship Phased NAWA F2 deployment under assessment

3 in diligence · 3 under review

Two sides

The arrangement has two counterparties.

Get compute