Platform
600 MW of AI compute the grid can schedule.
Project Ohm builds modular compute nodes and the platform that orchestrates them. It runs our nodes and our partners' as one fleet and is designed to take on third-party compute as well. To the grid, all of it is demand that rises with the midday surplus and steps back at the peak.
600 MW: Project Ohm rollout plan, September 2026
00.2The opportunity
AI load is coming either way. Ours moves with the grid.
< 0MW
minimum demand, forecast
Midday needs load.
AEMO forecasts minimum demand on WA's grid, the SWIS, to fall below zero within ten years. Demand that turns up at midday is exactly what the system is short of.
AEMO, 2026 WEM Electricity Statement of Opportunities (opens in a new tab)
296MW
curtailed on average, Q1 2026
Clean power is going to waste.
Network curtailment of wind and solar in the National Electricity Market (NEM) averaged 296 MW in Q1 2026, nearly double a year earlier. That's power waiting for a load in the right place.
AEMO, Quarterly Energy Dynamics Q1 2026 (opens in a new tab)
84MW
of midday demand contracted
Midday demand already has a market.
AEMO contracted up to 84 MW of minimum demand services in WA, and Western Power went to market for demand in Geraldton's solar hours. Networks pay for the kind of demand the fleet is built to provide.
AEMO, minimum demand services tender, 2023-24 (opens in a new tab)Western Power, tender WS5313524424 (opens in a new tab)
67GW
proposed data centre connections, up from 38 GW in a year
New load is arriving fast, and rigid.
Proposed data centre connections in the NEM went from 38 GW to 67 GW in a year, and 9.0 GW of large data centres are already in AEMO's transmission connection process. AEMO expects data centres to run as relatively inflexible loads. Ours is built to do the opposite.
AEMO, 2026 Electricity Statement of Opportunities, p. 37 (opens in a new tab)AEMO, Quarterly Energy Dynamics Q2 2026, p. 43 (opens in a new tab)AEMO, Digital demand surge (opens in a new tab)
The government proposes that large data centres reduce power when the grid needs it. It has been our design from day one.
Prime Minister, AI in Australia's interests, 15 July 2026 (opens in a new tab)
00.3The node
Modular compute at the substation.
Our nodes are modular compute units of 1 to 20 MW, built with tier-1 hardware partners. They sit at or near distribution substations and utility sites, and at renewable generation sites.
1 to 20 MW: Project Ohm rollout plan, September 2026
01Grid connection
Connects the node to the substation's switchyard. A small battery, not a diesel or gas generator, gives running jobs time to save their progress if the power drops.
02Metering and telemetry
Measures the node's power and streams live operating data to the orchestration platform.
03Compute array
High-density accelerator racks from any vendor, for AI training, fine-tuning, and batch and live inference.
04Cooling
Air or closed-loop liquid cooling that carries the heat away from the racks, designed to avoid the ongoing water use of evaporative cooling.
00.4The platform
Multi-variable scheduling for dynamic electricity grids.
Our nodes run flexible and time-critical AI work alike. Flexible jobs, such as training, fine-tuning, batch inference, embeddings, synthetic data, simulation and rendering, follow the cheapest and most abundant power. Time-critical work, such as live inference, runs on the node closest to its users. The orchestration platform decides when and where each job runs, and it is hardware-agnostic by design, so each job can run on the accelerator that suits it best. It weighs four things:
Energy availability
Electricity prices, renewable output, weather and demand forecasts, carbon intensity and grid constraints: how much power there is, where, and at what cost.
Job urgency
Each workload's requirements, from live inference that has to answer at once to a batch job with a deadline hours or days away. The more room a job has before its deadline, the more often it can run on cheap power.
Latency
Network conditions and compute availability: how close a job must run to its users and its data, how long the data takes to move, and the cost of loading a model onto another node.
Data security and sovereignty
Where each job's data is allowed to be, private networking, and each customer's compliance rules. Progress is saved to encrypted storage, and results go back to the customer's own storage.
And from them, it decides:
- Where workloads run.
- When they run.
- How capacity responds.
00.5On the grid
Connected like a load. Dispatched like a resource.
Every node sits behind the meter. The orchestration platform decides when each one runs and how hard, following the electricity market as it moves: live data, network constraint forecasts and regional supply. So the whole fleet acts as one resource the grid can schedule: a Dispatchable Compute Resource.
- Spot price, $/MWh
- Compute, simulated
Soaks up the surplus.
Ramps up when solar and wind outrun demand, adding load right where the network needs it.
Steps back on cue.
Eases off at the evening peak, when households need the power most, or when a feeder is stretched. The work waits or runs somewhere else, and jobs save their progress as they run, so a pause loses nothing.
Takes instructions.
Built to follow a network or market signal, not just price.
Shows its working.
Every node is metered and monitored, so network operators see what it drew, when, and how it responded.
Plugs into today's markets.
A Dispatchable Compute Resource is a new category, built to fit the mechanisms networks and markets already use.
In WA
Network support services, the Demand Side Programme, essential system services and Scheduled Facility registration.
In the NEM
Wholesale demand response, frequency control ancillary services (FCAS) and voluntary scheduling.
We tested the scheduler in simulation against 21 days of real five-minute AEMO prices.
00.6In the community
Built to work with communities, not against them.
Project Ohm is designed to earn social licence through existing infrastructure locations, low-water cooling and measured contributions to the electricity transition, not by asking communities to accommodate a new generation of data-centre campuses.
Using land that's already there.
We locate our nodes at or near existing substations and utility sites, reducing the need for large new developments. They connect to the network that is already there, with no new high-voltage lines.
Making better use of renewable energy.
Our nodes increase activity when renewable energy is plentiful and reduce demand when the network is under pressure, so households come first at the evening peak.
Using less water.
We use air cooling or closed-loop liquid cooling, designed to avoid the large ongoing water use of evaporative cooling.
No diesel or gas backup.
Because the work can move, our nodes are designed without diesel or gas generators. A small battery gives running jobs time to save their progress instead.
Smaller, modular nodes.
Our nodes take up less space and have less visual impact. They can grow as needed, and can be relocated if circumstances change.
Exploring new technology.
We work with technology partners to explore thermal batteries, heat recovery and advanced cooling that cut energy waste.
Investment in hundreds of towns.
Our Australian roadmap targets 300 to 400 substations, so the investment reaches hundreds of communities, not one campus.
300 to 400 substations: Project Ohm rollout plan, September 2026
00.7Carbon receipts
Time-matched carbon accounting.
Enterprise AI demands verifiable environmental reporting. Every node meters its own power, and the platform knows where and when each job ran, so we are building carbon receipts, a time-stamped record of each job: where and when it ran, and the emissions of the power it drew at that hour.
Carbon receipt
- Job
- Where
- When
- Emissions of that hour's power
00.8The rollout
From one substation to hundreds.
It starts with Node 01, in a substation in Geraldton, WA, with Western Power, and a second site is in planning on the east-coast market. From there the fleet grows across Australia's two main grids, the SWIS and the NEM, to 300 to 400 substations, each node sized to the capacity on site and installed on a nine-month target deployment cycle, using network capacity that already exists, with XENON supplying the hardware.
At full scale, that's a power station's worth of demand the grid can turn up and down.
Built to move.
When a substation no longer needs the extra load, its node goes on a truck to one that does. A fixed data centre stays where it was built; this fleet goes where the grid needs it.
Grows faster than we can build.
The fleet grows two ways: nodes Project Ohm funds itself, and nodes funded with partners, from utilities and energy partners to infrastructure investors. Beyond the fleet, the orchestration platform is designed to take on third-party compute and schedule it the same way: 600 MW in the fleet, and ultimately gigawatts under orchestration.
600MWtarget fleet
300+substations
1MW+at each site
- Now Node 01, in Geraldton
- Next the east-coast market
- Then 300 to 400 substations across the SWIS and the NEM
- After that Southeast Asia, then beyond
600 MW across 300 to 400 substations, each from about 1 MW: Project Ohm rollout plan, September 2026
Be on the map.
Substations with spare capacity at midday, and wind and solar farms that are curtailed, are where the fleet goes next. We're choosing the first sites now.