DEPLOYMENT / ONE PLATFORM, FOUR FORMATS

From cell tower to multi-gigawatt.

Training workloads cluster at gigawatt campuses. Inference lives at the edge — cell towers, hospitals, telco co-los, sovereign nodes. The compute market is splitting in two directions, and the infrastructure underneath has to follow. We build for both. One modular foundation. Four physical formats. Same delivery discipline at every scale.

01 Edge 02 Distributed 03 Campus 04 Multi-GW
01 / EDGE < 1 MW

Edge nodes.

Inference compute belongs where the users are. Cell towers, hospitals, retail anchors, telco co-los, government facilities. Sub-megawatt modular pods deployed in distributed fleets — locking the layer of infrastructure that almost no one is building for yet.

THE THESIS

Training is hyperscale. Inference is everywhere. As AI moves from model development to widespread consumption, the compute layer has to follow the user — and ultra-low-latency delivery requires infrastructure that lives at the edge of the network, not in a gigawatt campus three states away.

We built our platform for both halves of that picture. The same factory that builds our campus pods builds the sub-megawatt pods that drop into cell tower compounds, hospital backlots, and telco facilities. The distributed edge layer that inference economics actually require is something we manufacture, deliver, and operate today — not a future workstream.

Hundreds of megawatts of distributed capacity across thousands of edge sites — a network nobody is building yet. The next layer of the AI compute stack.

CELL TOWERS
Telco co-locations
Sub-megawatt pods at carrier mast sites. Inference compute delivered with sub-millisecond latency to the user, no fixed-line dependency.
HOSPITALS
Healthcare AI
Imaging, diagnostics, clinical decision support — running on-premise behind the firewall. Compliance by design, not retrofit.
SOVEREIGN
Government nodes
Controlled-access edge compute for federal and defense workloads. Modular footprint accommodates secure facilities and contested logistics.
RETAIL · ENTERPRISE
Distributed anchors
Anchor tenants needing dedicated regional inference — retail chains, regional banks, telco resellers. Multi-site, single operating model.
02 / DISTRIBUTED 1–10 MW

Distributed sites.

Multi-site clusters in the 1–10 MW range. Regional inference deployments, enterprise pilots, federated training across geographies. The pod count scales up, the deployment discipline stays identical to the edge.

THE FIT

Between edge nodes and full campuses sits the distributed tier — multi-site programs where regional latency, sovereignty boundaries, or anchor-tenant economics call for several million-watt sites instead of one big one. Enterprise AI pilots, federated regional inference networks, and sovereign workload portfolios all live here.

Same modular factory. Same delivery discipline. The customer gets a fleet of sites, not a single point of capacity — with operational uniformity that single-site builds can't match.

REGIONAL INFERENCE
Multi-metro coverage
2–8 sites across major metro regions, identical operational footprint. Inference workloads served from the metro they're consumed in.
ENTERPRISE PILOT
5 MW proof deployments
Pre-commitment pilot capacity for hyperscaler or enterprise customers. Proves the model on the customer's actual workloads before scaling.
SOVEREIGN FEDERATED
Multi-jurisdiction
Distributed compute across regulatory boundaries — data residency, export controls, government partner requirements all addressed at the site level.
BROWNFIELD CONVERSION
Industrial site reuse
Modular pods enable deployment on brownfield and former industrial sites. No ground disturbance required — sites traditional DC builders can't use.
03 / CAMPUS 10–300 MW

Single-site campus.

Behind-the-meter generation on dedicated acreage. Training-grade workloads. Single anchor tenant or anchor-plus-multi structure. The model the market knows — built with modular discipline instead of stick-built risk.

THE FORMAT

Where dedicated training capacity matters — multi-hundred-megawatt single-site deployments serving anchor tenants requiring concentrated GPU clusters. The pod-based delivery model means capacity activates in waves rather than waiting for the whole site to commission, and refresh cycles compound the long-term value.

Behind-the-meter generation eliminates the interconnect-queue exposure that defines every traditional grid-tied campus build right now. The site is operational on company-owned power infrastructure while utility tie-in proceeds independently.

HYPERSCALE TRAINING
Anchor-tenant campus
Single offtaker, dedicated multi-hundred-MW deployment. Behind-the-meter primary, grid pursued in parallel. Refresh model preserves value over the lease.
NEOCLOUD
GPU-cloud capacity
Dedicated capacity for AI-native cloud operators. Credit structures built around the offtaker reality, not theoretical IG-grade assumptions.
MULTI-TENANT
Anchor + adjacency
Anchor tenant takes the majority footprint, adjacency capacity offered to secondary tenants. Same modular shell, partitioned by pod.
REFRESH-AWARE
10-year + 2 swaps
Lease structure assumes two hardware refreshes inside the term. Customer locks in long-term capacity without locking in obsolete infrastructure.
04 / MULTI-GW 300+ MW

Multi-gigawatt programs.

Federated multi-site programs at gigawatt scale. National platforms, sovereign capacity portfolios, hyperscaler partnership programs. One operating system, many physical sites, delivered as a coordinated program rather than a one-off build.

THE PROGRAM

At the largest scale, customers are no longer buying a building. They're buying a program — multi-gigawatt capacity across multiple sites, multiple regions, coordinated under a single operating model. This is where the modular thesis becomes unanswerable: no traditional builder can deliver gigawatts of distributed capacity on a coherent platform. The model that builds one site builds all of them.

Site selection, manufacturing throughput, power partner relationships, and capital structure all scale linearly. The 300th megawatt deploys the same way the first one did — with the same delivery discipline, the same refresh model, the same operational uniformity.

NATIONAL PLATFORM
Multi-region program
Multi-GW capacity across multiple US regions. One platform, one operating model. Customer signs once, scales across the country.
SOVEREIGN AT SCALE
Government compute
Multi-site sovereign capacity portfolio — federal, defense, and adjacent agency workloads with controlled-access, controlled-supply-chain delivery.
HYPERSCALER ANCHOR
Strategic partnership
Hyperscaler partnership structures where Fullscale operates dedicated multi-GW capacity under long-term agreement. The deliverability the customer can't build internally fast enough.
EXPORTABLE MODEL
Replicable abroad
The platform deploys identically in any jurisdiction where modular, behind-the-meter capacity is the right answer. Canada, allied markets, sovereign-friendly geographies.
THE COMPUTE SPLIT

Training centralizes.
Inference distributes.

Model training concentrates compute at gigawatt campuses where the cluster economics and chip-to-chip bandwidth demand a single location. Inference does the opposite — it follows the user, requires sub-millisecond latency, and scales to wherever people actually consume AI products. We built the platform to serve both.

The same modular pod design deploys at sub-megawatt cell-tower compounds and at multi-hundred-megawatt campuses. One platform, both halves of the picture, the same delivery discipline at every scale.

TRAINING
Cluster centralized.
  • Single anchor site
  • Gigawatt-scale concentration
  • High inter-chip bandwidth
  • Behind-the-meter power critical
  • Latency to user not material
FULLSCALE FORMAT · CAMPUS · MULTI-GW
INFERENCE
Network distributed.
  • Many small sites
  • Sub-megawatt to 10 MW
  • Latency to user is the metric
  • Lives at network edge
  • Where Fullscale's distributed format fits
FULLSCALE FORMAT · EDGE · DISTRIBUTED
CAPACITY OPEN

What format fits your workload?

Submit a capacity inquiry. Tell us the workload, the scale, the timing. We respond with a fit assessment.

Submit capacity inquiry