Many small sites.
One AI network.
Vulcore builds distributed AI infrastructure: a network of battery-backed compute modules behind the meters of commercial buildings with spare electrical capacity. Each Victor module is colocation for your AI servers, with power, cooling, a battery and a locked site in one box, running on the building's existing service with no new grid connection.

Distributed AI infrastructure
Many small sites, run as one network.
Most new AI capacity is planned as giant campuses, and each one waits for a new grid connection. In the US, 2,060 GW of projects are in line to connect, and the median wait is five years, according to Lawrence Berkeley National Laboratory.
Distributed AI infrastructure takes the other road: many small sites, each on power that's already delivered, linked into one network. It suits the way AI is used. Training a frontier model takes thousands of chips in one building. Serving it doesn't: each copy runs on one server, and each request can go to the closest site. A robot, a car or a camera can't wait on a campus three states away.
AI compute behind the meter
Why the battery is the interconnect.
Behind the meter means on the building's side of its electric meter, on service the building already has. A Victor module is a regular commercial account that never sends power back to the grid, so it gets a utility work order instead of a place in the line to connect.
The battery is what makes that work. It keeps the building and the module together inside the service that's already there, and it carries your servers for hours if the grid goes down. Qualifying is the work: a site visit checks the transformer, the wiring and the space before anything is signed.
How it compares
The other ways to get AI servers running, and where a Victor fits.
- Hyperscale campuses. Hundreds of megawatts on one site, behind a new grid connection that takes years. The right place to train frontier models. A Victor module is about a tenth of a megawatt, and it's live in weeks.
- Modular data centers. Factory-built rooms that ship quickly but still need a site with power, which is now the scarce part. A Victor module is factory-built too, and it brings its own battery to a building that already has the power.
- Colocation in existing data centers. The right answer where space and power are open. Vulcore adds capacity where they aren't: in commercial buildings with spare service, close to the people the work serves.
- On-site generation. Turbines, engines or fuel cells built next to the load. It makes new power, and it brings fuel, air permits and noise with it. Vulcore's commercial sites run on the building's service and the battery, with no generator on site.
- GPU clouds. You rent hours on someone else's machines. Vulcore is the other half: you bring your servers, and we power, cool and protect them. We don't sell GPU time.

Where it fits in the AI buildout
The data center supply chain, on-premises AI, and companies that own their models.
- The data center supply chain. Transformers, switchgear and new grid connections are the slowest parts of adding AI capacity, and the gear is back-ordered for years. At a site that qualifies, a Victor module needs none of them: it runs on the service the building already has, and its battery, power conversion and cooling are built in at the factory.
- On-premises and air-gapped AI. Compute on your own property, inside your own fence, with no shared hall and no shared network. Work that never touches the public internet fits, for agencies, banks, hospitals and defense sites.
- Companies that own their AI. For companies that run their own models on their own servers instead of renting from a cloud. A Victor module gives that hardware power, cooling, a battery and a locked site, next to the company's data. Your servers, your software, your keys.
- Manufactured, not developed. Every Victor is the same manufactured design, built once and deployed everywhere. A new site is a copy of the last one, not a construction project.
Who it's for
Compute partners, companies running their own AI, property hosts and the public sector.
- Compute partners. Neoclouds, AI labs and enterprises with servers waiting on power. One module, or a megawatt across a metro under one contract. See the use cases.
- Companies running their own models. Banks, hospitals, manufacturers and software companies that want their AI on their own hardware, next to their own data. Your servers, your software, your keys.
- Property hosts. Owners of warehouses, light industrial, retail, self-storage and fast-charging sites. Rent on electrical capacity you already pay for and don't use. We finance, install, insure and run the module.
- Defense and the public sector. Installations, agencies and critical public facilities that need compute inside their own fence. See ruggedized and public sector.
The Victor class
One module, built at three densities.
- V-30. One air-cooled rack at 30 kW on a 10-foot skid, with a 150 kWh battery.
- V-60. Two racks at 60 kW, air or liquid cooled, in a 20-foot container, with a 300 kWh battery.
- V-100. Three to four racks on direct liquid cooling at 100 kW in a 40-foot container, with a 520 kWh battery.
Ruggedized and public sector
The V-100R, and AI compute that stays on public property.
- The V-100R. The ruggedized Victor: the V-100's 520 kWh battery and 100 kW of compute in a hardened 40-foot container that moves by truck, rail or sea. It runs on grid power, tactical generators, vehicle power or deployable solar, so the same module works at a fixed installation or a forward site.
- A ruggedized family. Smaller V-30R and V-60R modules for sensor sites, communications nodes and command posts, and an air-deployable version for cargo aircraft, all on the same electrical interface. Details on the defense page.
- Federal, state and local agencies. Compute on public property, inside your own fence. Your servers and your keys, machines in the United States, and data that never leaves the site. Air-gapped work fits.
- Critical public facilities. Utilities, water systems, public hospitals, schools and universities. The battery keeps critical circuits on when the grid goes down, and the module runs AI on capacity the building already has.
- Supply chain and ownership. Every part is screened against federal foreign-entity (FEOC) restrictions. Vulcore finances, owns and operates the module.

Where and when
Texas first.
The first sites are planned in Texas for 2027. Pricing is one monthly rate per kilowatt reserved, with power, cooling, the battery and the site included, quoted per site.
The FAQ answers the rest: power, safety, water, hosting and compute.