Your home can power AI.
A small unit in your side yard does AI work, using power your home isn't.
Every home runs a node.
Linked together, they work as one data centre.
1 Why
AI needs more power than the grid can add.
AI demand is growing 10× a year. Data centres here grow 1.3×. Meanwhile, about half of the power a home is connected to goes unused off-peak.
10×/yrAI demand
1.3×/yrGCC data centres
~50%idle off-peak
Why now. Sovereign-AI mandates, fixed demand-charge power, GPU availability and new Vision-2030 communities all line up. $30B is committed to GCC data centres, with 200 MW live by '27.
Chart shape is schematic.
2 How it works
Thousands of homes instead of one huge building.
Before: 200 MW in one campus.
New land~40–80 hectares0
Build time18–24 months1 day per node
Grid connection4+ yearsAlready at the home
Building cost~$1.8B, before GPUs$0, no building
One failure takes outA whole site at stake1 home in ~32,800
3 The node
A sealed unit, installed in a day.
Full specs
4× RTX PRO 6000Server Edition, 500 W avg, 600 W max
2× dual-socket servers92% PSU efficiency
2× 4 kW AC (N+1)COP 1.5 at 45°C, closed refrigerant loop
6.1 kW meteredIT 3.4 kW + AC 2.7 kW
1900 × 950 × 1300 mm26RU, 19" industrial rack
IP55 / NEMA 4IP68 bottom cable entry
6 KVA UPSDual metered PDUs (A / B)
Switch A + BPrimary and high availability
Service accessFront, rear and bottom
Sealed re-entryKeeps the IP rating after service
Power and network side
316 stainless steelInsulated double wall, pad-mount plinth
IP66 / NEMA 4IP66 cable glands, rack rails, busbar
6 kVA / 6 kW UPSOnline double conversion, 20 × 7Ah, up to 4× in parallel
4,000 W, 230 V ACCooling, heater and dehumidifier
Smart PDU8 outlets, remote outlet control, overload protection
Sensor portsTemperature, humidity, optional SNMP
4 The network
Homes share the work between them.
- You askYour request lands at the nearest home.
- Neighbours pick it upIt spreads home to home. Nothing sits in the middle.
- Everyone does a partEach home runs a piece of the job.
- One drops out, nothing stopsIts neighbours take over the work.
- You get one answerThe pieces come back through the homes to you.
Home workingOffline, neighbours cover
6 Our offering
Two ways to join: one node, or a Pod.
A sealed unit beside your villa. Installed in a day, running on power your home isn't using.
4GPUs
384 GBGPU memory
1 dayto install
A compact, self-contained module. It ships quickly, installs in weeks, and joins the mesh anywhere power is available.
72GPUs
~10 TBGPU memory
2–4 wksto deploy
All Pod specs
H100 / H200 / B200GPU options, configs in validation
~10 TB HBMH200 shown, 5.8–13.8 TB by config
NVLink / NVSwitchInside the module
400 Gb/sIB / RoCE, mesh-ready
Scaling unit+1 Pod adds 72 GPUs
One nodeDrops into the Community mesh as one node
SealedSecure boot and tamper detection
Keys wipedOn physical intrusion
Compare side by side
| Single node | Pod | |
|---|---|---|
| GPUs | 4 | 72 |
| GPU memory | 384 GB | ~10 TB |
| GPU-hours a year | 35,040 | 630,720 |
| Power | 6.1 kW | ~90–110 kW |
| Deploy | 1 day | 2–4 weeks |
| Footprint | 1.8 m² | ISO module |
| Cooling | Closed refrigerant loop | Sealed liquid loop + free air |
| Where | Beside a villa | Anywhere power is available |
5 Security
Nobody at the site can see the data.
Every job stays encrypted while it runs, and every node must prove it is genuine before it can join.
Confidential computingHardware TEEs. Site operators never see the data.
Encryption everywhereAES-256 at rest, TLS 1.3 in transit, encrypted links.
Node attestationFirmware and hardware checked. Fail, and it's quarantined.
Tenant isolationNo GPU memory is ever shared.
Zero-trust fabricSigned identity and mutual TLS, even on one site.
Customer-managed keysHSM-backed. Revoke access instantly.
Tamper resistanceSecure boot. Keys wiped on intrusion.
Data residencyPin workloads to a country or site.
7 Community
Every home that joins adds to the network.
Tap any home to add it. Each one links to its neighbours.
~50KUAE villas
~300KVision-2030 units
How homes link up
Devices join one elastic fabric. Compute is pooled, scheduled and load-balanced, and the network grows with each device that joins.
512 GPUsScale across the mesh
400 Gb/sIB / RoCE uplink, NVLink / NVSwitch
H100 / H200 / B200GPU options in validation
Up to 50°CStandalone, outdoor rated, liquid-to-air
8 Order yours
Host a node, or back the network.
Illustrative simulation
