Pods put up to 160 GPUs on power a site already has
Modular edge data centers in two sizes, 125 kW and 250 kW, plus a mobile Battery Pod. They sit in a few parking spaces, connect behind the host's existing service and move when the site's needs change.
Meet the Pods
Each pod is a complete edge data center in a standard container: GPU servers, networking, cooling, power distribution and controls. A Battery Pod beside it covers peaks and grid events.
Key features
- Modular and mobile
Standard 20 ft and 40 ft containers that move to a new host when load or financing changes.
- Runs on the existing service
Modulo Site AI holds the import limit, so a site that's ready now needs no new service.
- Serves offline
Each pod runs its own Kubernetes with models cached on site, so it keeps answering when the link drops.
Engineered for spare-capacity sites
- Pod 125: power for IdlePower equipment
- 125 kW
- Pod 125: GPUs
- Up to 80, H100-class, in 10 servers of 8
- Pod 250: power for IdlePower equipment
- 250 kW
- Pod 250: GPUs
- Up to 160, H100-class, in 20 servers of 8
- Model serving
- KServe with vLLM and Triton
- Connection
- Behind the host's existing service
- Limit held by
- A UL 3141 power control system, or an energy management system under NEC 220.70
- Power inputs
- Grid, solar, storage and local generation
- Firm share
- Up to 80% of the site's power envelope
- Rating
- 250 kW / 1,000 kWh
- Smaller unit
- 250 kWh, towable behind a pickup
- Listing
- UL 9540 battery systems
- Moves
- By season, never daily, with a licensed driver
- Footprint
- A few parking spaces, plus a battery pad with 10 ft clearances
- Kubernetes
- K3s on Kairos, on every pod
- Offline
- Models cached on site with Dragonfly; serving continues without the link
- Cloud consoles
- Each pod can join the Kubernetes console you already use
GPU counts scale a reference pod of 48 H100 GPUs at 75 kW.
Pod configurations
Pick the size the site's headroom allows. Most sites start with one pod and one battery.
Pod 125
Up to 80 GPUs for sites with room for one smaller pod. The usual starting point.
Pod 250
Up to 160 GPUs for sites with more headroom, such as logistics parks and plants.
Battery Pod
Covers peaks and grid events, and moves between hosts by season.
One controller holds the whole site to its limit
Modulo Site AI sits between the meter and everything behind it: the host's own loads, the battery and the GPU pod. The pod never pushes the site past the service it already has.
From site check to a live pod
Modulo Digital Twin does the first pass from a desk. The rest follows the same five steps at every site.
- 01
Site check
Bills, photos and Modulo Digital Twin's first read.
- 02
Design
Pad, service connection and clearances.
- 03
Permits and limits
Fire permit and any utility limit agreement.
- 04
Delivery
Pod and battery set and connected.
- 05
Live
The Modulo Orchestrator takes over.
What the site provides
- Room for a pod in a few parking spaces
- A battery pad with 10 ft clearances
- Spare capacity on its existing service
- Fiber or another reliable network link
- Truck access for delivery
What IdlePower delivers
- The pod, the Battery Pod and Modulo Site AI
- Installation through a local electrical contractor
- Monitoring and operations
- Settlement from the revenue meter
Designed to the rules for batteries and service capacity
Mobile batteries and added load on an existing service both have clear rules in California. We design every site to them.
- California Fire Code 1207.10
- Mobile battery permits, 10 ft clearances and stay limits
- UL 9540
- Listed battery systems
- UL 3141 or NEC 220.70
- The control system that holds the import limit
- Rule 21
- Interconnection for grid-parallel batteries at each dock
- DOT hazmat (UN3536)
- Whenever a battery moves by road