Vera Rubin-ready
Designed for next-generation AI platforms: the power density, DC architecture and liquid cooling they call for.
DG Energy & Infrastructure
Purpose-built
for AI workloads.
An on-site power plant. Built for AI today and the demands of tomorrow.
Data centers built close to the customer, on land that fits the community.
Designed for existing industrial buildings and new greenfield developments. Initial behind-the-meter operation does not depend on a utility interconnection queue.
See how it works Talk to us →Conceptual design. Block availability is stated as net IT capacity; generation figures are labeled separately. Subject to change, not operating performance.
- 10 MW blocks across seven states
- 553
- Conceptual net IT availability
- 5.53 GW
- Rack design specifications
- 180 kW AC/DC, scaling up to 480 kW DC
- Power generation
- On-site
Each block delivers 10 MW of net IT. Availability is shown by state and is conceptual and subject to change.
What sets us apart
AI is the starting point.
Twelve reasons the power, building and cooling work as one system. We did not adapt an enterprise data center.
Fast deployment
A standardized, factory-built design repeated at every site, with delivery subject to project milestones.
Behind-the-meter power
On-site solid-oxide fuel cells generate power where AI compute runs.
Fuel cells, not engines
Fuel cells convert fuel to electricity electrochemically. Project-specific equipment and permitting requirements still apply.
No long waits
Dedicated on-site generation reduces reliance on utility interconnection. Grid-fed portions of hybrid campuses still require utility service.
Standardized distribution
One engineered power distribution standard from the fuel cell to the rack, repeated in every building.
Repeatable buildings
One standardized H building across many sites. Design once, deploy everywhere.
Waterless liquid cooling
Direct-to-chip liquid cooling with dry heat rejection instead of evaporative cooling. This does not mean the entire campus uses no water.
Scales in 10 MW steps
Add net IT capacity one block at a time, from a single block to multi-gigawatt campuses.
Standard compute pod
About 3.34 MW of IT and 16 compute positions: the same building block at every site.
Rack density for AI
180 kW AC/DC, scaling up to 480 kW DC. Design specifications, not measured performance.
AC or DC: Cassette Swap
Run AC, DC or a mix in the same building, changing pod by pod as AI hardware evolves.
The constraint
AI demand is
outrunning the grid.
New large loads face multi-year waits. In Northern Virginia, Dominion estimates roughly four to seven years for a new large data center connection.
Why the system is broken →4–7years waiting for a connection¹Read the source
Our answer
We are the power plant at the rack.
Independent 10 MW net IT blocks put generation beside the compute. Each block has its own fuel cells, supercapacitors, battery ride-through and cooling.
Standardized distribution, swappable AC or DC cassettes and direct-to-chip cooling with dry heat rejection connect the gas meter to the AI rack.
Explore our solution →Campus tour
From power plant to compute.
Read the tour description
On-site generation, compute space and direct-to-chip liquid cooling. The H building joins two compute legs through a central connector.
The film moves from an aerial view of the H building to fuel-cell equipment, distribution, cooling, compute pods and racks. Independent blocks deliver 10 MW net IT; generation capacity is separate from net IT capacity.
180 kW AC/DC, scaling up to 480 kW DC. These are design specifications, not measured performance.
Dry heat rejection avoids evaporative cooling. Other water needs and local impacts require site-specific validation. Repeated buildings illustrate a design concept, not deployed capacity or the state inventory.
We are the power plant at the rack. Conceptual and subject to design development; no guaranteed site outcomes.
Independent blocks
Redundancy in isolation.
Each 10 MW block is designed with its own spare power and cooling, without a shared electrical bus or shared cooling between blocks. Medium-voltage distribution places the power plant approximately 200 feet from the GPUs in the illustrated design.
Initial behind-the-meter operation does not depend on a utility interconnection queue. A grid tie can add support and an outlet for export, subject to utility agreements and local approvals.
If a tenant's needs change, the generation system is designed to serve another tenant or a different load. Export is an option only where approved. Manufactured generation and cooling equipment is designed for redeployment where site conditions permit.
Design intent, not demonstrated fault containment or operating availability. Block separation does not establish independence of upstream fuel supply or utilities. Grid ties, export and equipment redeployment require project-specific validation and approvals.
Beyond AI
Power plants for AI-driven industry.
Across existing industrial buildings and new greenfield developments, blocks designed for AI compute can serve other demanding loads. The power plant is an adaptable asset for AI-driven industrialization. Compute pods and distribution are designed for compute redundancy and ride-through; robotics, automated manufacturing and cold storage require configurations matched to their own loads.
Robotics
On-site generation can support robot fleets and the facilities that build and test them.
Advanced manufacturing and automation
On-site generation can support automated production lines in existing or newly developed facilities.
Cold storage and automated logistics
Power made on site can support refrigeration and warehouse automation.
Industrial electrification
Blocks can add capacity in steps as industrial processes move to electric power.
Grid support
An approved grid tie can enable power export where the utility and local approvals allow.
Secure government use
Dedicated power can support public-sector workloads. Security requirements are scoped project by project.
A power plant. Many possible uses.
Potential uses shown for illustration. Conversion is not guaranteed: distribution, controls and cooling must suit the new load. Grid resale requires agreements and approvals; no resale yield, residual value or underwriting outcome is established.
Speed to power
Years in line. Or a standard you can repeat.
Waiting on the grid 4–7 years¹
- Large-load request & queue position
- Load studies & cost allocation
- Transmission & substation upgrades
- Energization, if capacity exists
The DGEI standard No interconnection queue
- Site, fuel supply & permits
- Factory-built blocks and H building, in parallel
- Commission capacity block by block
AI compute live, one 10 MW net IT block at a time →
Illustrative; not to scale. This comparison describes dedicated on-site generation. Grid-fed capacity, grid ties and export require utility service or agreements and approvals. Delivery is conceptual and subject to development milestones; no delivery date is promised. ¹ Dominion estimate for Northern Virginia large-load connections; source 1.
Built to scale
One H building.
A repeatable design.
Meet your AI guide
Your questions. A conversation.
Explore on-site power, the H building and AI cooling with our interactive guide.
Meet the DGEI guide →