smartgrid.com.au is building a new kind of AI infrastructure for Australia. Compact compute nodes beside homes — and filled racks inside the server & comms rooms of embedded networks, airports, shopping centres and other managed facilities. Powered, networked and physically secured on day one.
Every powered comms room is a potential node.
AEMO forecasts data centre electricity demand rising from ~4 TWh today to ~21 TWh by 2034–35 in the NEM — roughly 9% of demand.
22 state-significant data centre proposals approved in NSW. The NSW IDA endorsed $51.9B of further proposals in March 2026.
A greenfield data centre takes years to plan, approve, grid-connect and commission. Powered, secured comms rooms already exist today — and sit largely idle.
Most suburban distribution networks run at under half their capacity. That headroom is the cheapest gigawatt in the country.
Australia doesn't need to choose between an AI economy and an affordable grid.
The cheapest gigawatt is the one already paid for — sitting unused in the wires under suburban streets.
— smartgrid.com.au · Program brief, May 2026
We deploy pre-integrated GPU racks straight into the server & comms rooms of facilities that already run powered, networked and access-controlled infrastructure. No new building, no land, no multi-year approval — the hardest parts are already done.
Where a facility isn't nearby, a compact node sits beside a home — paid for with a free battery, cheaper power and a share of the inference economy. Distributed capacity, close to where users actually are.
A CEC-approved smart panel, a 10–13 kWh home battery, and a compact compute node beside the house. Installed by SAA-accredited electricians under AS/NZS 3000 and 4777.
Liquid-cooled GPUs in a weatherproof enclosure roughly the size of an outdoor AC unit. Designed for under 35 dB at one metre. Slab-mounted, optionally fenced.
Orchestration runs AI inference workloads only when the local network has spare capacity. Your battery always covers your home's load first.
Choose monthly bill credits, a flat $0 energy + internet package, or take cash. Plus real backup power during outages — independent of the grid.
If your site already has a server or comms room — power, redundant internet and a real security posture — most of what a data centre needs is already in place. We bring pre-integrated, filled compute racks and the offtake; you contribute space that's largely sitting idle. This is the faster, denser, more secure path, and the one we lead with.
A site survey scores your comms/server room against four constraints — power, cooling, network and security. We're upfront: a room rated for telco or IT gear can't always take GPU-rack density without a cooling retrofit, so we confirm thermal headroom before committing.
Pre-integrated GPU racks roll into your existing rack space, behind your existing access control. Liquid or rear-door cooling where density demands it. No structural works, no new perimeter — we use the building you already run.
A dedicated, separately-metered feed — and in embedded networks, a child load behind your parent connection where there's headroom. We pay for every kilowatt the racks draw. Your operational load is never touched.
Recurring revenue, an energy offset, or a share of the upside — your choice — from space and capacity that previously earned nothing. We monitor, maintain and operate the fleet remotely.
"Has a server room" isn't the same as "can host AI compute." Here's exactly what we check — and what makes a site a strong candidate.
Dedicated circuit capacity, three-phase supply, UPS runtime and ideally generator backup. We model spare capacity before we add a single rack.
The constraint that decides most sites. Rooms built for ~3–8 kW/rack need rear-door heat exchangers or liquid cooling to take modern GPU density. We design to your headroom, never assume it.
Commercial-grade, ideally diverse-path fibre with low latency to the workloads we serve. The racks run on their own connection, isolated from your operations.
Your existing posture is the selling point: controlled access, CCTV, monitoring and a clear chain of responsibility. We slot in behind it rather than rebuilding it.
Free rack U or floor tiles, with clearance for airflow and service access. A handful of racks is enough to start; we scale as headroom allows.
Where you operate an embedded network, we model the compute as a child load behind your parent connection and confirm headroom before the gate trips — with metering that keeps it cleanly separated.
Sensor fusion and vision inference for autonomous trucks, ports, mine sites and remote-operations centres — closer to where the trucks actually run.
Real-time defect detection, digital twins and condition monitoring for advanced manufacturing corridors and infrastructure operators.
Low-latency in-country inference for retailers, banks, government and the Australian AI startup scene. No data leaves the country.
GPU-backed sessions for cloud gaming and creative pipelines that benefit from being one suburb away rather than three continents.
Every node is a battery on the grid. Frequency response, peak-demand dispatch and wholesale arbitrage — automatically.
Burst capacity for university research, bushfire mapping, emergency-services analytics, and other work that can't wait for an offshore queue.
Phase 1 runs both tracks in parallel: filled racks into qualifying facility comms rooms — embedded networks and large managed sites first — alongside the first residential nodes in Western Sydney new-build estates. We then expand by site density and where federal and state incentives stack cleanly.
Tell us whether you operate a facility with a server/comms room, or want to host a node at home. Facility sites are prioritised for Phase 1. Registering takes a minute — we'll be in touch as each phase opens.
We're talking with hyperscalers, neo-clouds, AI startups, facility & embedded-network operators, distribution networks, retailers and developers. If any of the below sounds like you, get in touch — we move fast.
Partnership inquiries: partners@smartgrid.com.au
Sometimes, sometimes only after a cooling upgrade. Power and network are usually the easy part; the deciding constraint is heat. A room provisioned for ~3–8 kW per rack can't dissipate a dense GPU rack without rear-door heat exchangers or a liquid loop. Our site survey measures your actual thermal and power headroom first, and we only deploy what the room can genuinely support.
We can sit as a child load behind your parent connection where there's spare capacity, with separate metering so the compute energy is cleanly attributed and paid by us — never cross-subsidised by your occupants. Where headroom is tight, we bring a dedicated supply instead. We confirm the arrangement with you before anything is energised.
No. The racks run on their own dedicated connection and are logically isolated from your building systems. Workloads execute inside encrypted enclaves, so neither the compute customer nor Smart Grid can see your operations, and your team can't see theirs. Data stays in-country.
Provide qualifying space and existing services, and sign the site agreement. We fund and install the racks, any cooling upgrade and connectivity, then monitor, maintain and operate the fleet remotely. You're not responsible for uptime, workloads or customers — you receive recurring revenue, an energy offset, or a share of the upside.
The residential node is liquid-cooled and engineered for under 35 dB at one metre — roughly the sound of a quiet library. Heat is rejected via the liquid loop to a small radiator on the unit; we site nodes with airflow in mind, away from bedrooms wherever practical.
No. Whether at a home or a facility, the compute runs on a separate metered circuit and we pay for the energy it consumes. Your own energy use is unaffected. For home hosts, we also install a battery and smart panel at no cost, so most use less grid power than before.
The compute shuts down gracefully and its workload shifts to other nodes and racks on the network. At a facility, your own backup (UPS/generator) protects your operations as it always has. At a home, the battery powers your essential circuits until the grid returns.
Terms are set out in the pilot site or host agreement, including notice periods and removal of equipment at our cost. Home agreements are structured to transfer cleanly to a new owner with consent, or terminate with reasonable notice. Full terms are provided before you commit.
Phase 1 opens Q2 2027. Facility sites with existing power, network and security are prioritised — register yours, or join as a home host.