Data centres, industrial loads, EV hubs and microgrids

For a simplified constant 80 MW IT load and an assumed power-equivalent multiplier of 1.25, facility demand is 100 MW. At that constant level for 8,760 hours, annual energy is 876 GWh. This is an introductory estimate, not proof of a 100 MW peak connection requirement. Annual PUE cannot be applied mechanically to a real IT peak.

Connection planning needs time-resolved coincident cooling and IT loads, hot-weather cases, startup and restart ramps, harmonics and power-quality assessment, maintenance states and site phasing. A 20-to-100 MW rise over four minutes is 20 MW/minute. Whether that is acceptable is a network and control-study question, not a conclusion from annual energy totals.

A UPS bridges interruptions; standby generation supports longer outages; independent incoming supplies reduce some supply risks. None is a complete resilience solution alone. Map common dependencies such as the upstream substation, switchgear, controls, fuel delivery, cooling water and communications. Two feeders can still share a single point of failure.

If a protected 120 MW load needs 15 minutes of ride-through, delivered energy is 30 MWh. With 90% stored-to-load efficiency, the introductory stored-energy requirement is 33.3 MWh before reserve, ageing, temperature and reliability margins. Actual design also needs power capability, transfer timing, protection and specialist engineering review. This course is not a site operating procedure.

Energy cost follows facility demand, not server nameplates alone. A hypothetical 876,000 MWh annual facility load at an energy-only $80/MWh costs $70.08 million before network charges, losses, levies and contractual adjustments. Commissioning delays or a phased demand ramp can materially change both energy volumes and hedge requirements.

A solar PPA that matches annual consumption does not necessarily supply the same intervals or location as the load. Analyse interval residual demand, basis, credit, volume and commissioning risk. Keep physical supply reliability separate from financial hedging and environmental accounting: an annual renewable-energy claim is not evidence of uninterrupted renewable supply in every hour.

Some workloads can be delayed, shifted or curtailed; others cannot without breaching service commitments. Identify eligible computing tasks, notice times, maximum interruption, recovery demand and verification methods. Curtailing 10 MW for half an hour reduces energy by 5 MWh during that period, but deferred work may create a later rebound.

Efficiency also requires absolute-energy tracking. If IT energy doubles from 100 to 200 GWh while PUE improves from 1.4 to 1.2, facility energy rises from 140 to 240 GWh. A better ratio does not guarantee lower total consumption. Report energy, water and emissions claims with their own clearly defined metrics and sources.

The brief should reconcile phased IT capacity, expected utilisation, facility load profiles, network capacity and timing, connection costs, resilience architecture, power procurement, environmental claims and operating constraints. Record the owner and confidence of each assumption. Test hot weather, a delayed connection, slower customer uptake and an upstream outage.

Recommend a staged plan with decision gates: what must be proven before construction, energisation and the next capacity release? Identify the technical studies and contractual reviews still needed. Connect this work to energy units (Topic 1), procurement, large-load connection (Topic 41), physical resilience (Topic 58) and the integrated capstone (Topic 60).

A fleet charging plan begins with vehicles, routes, arrival times, departure deadlines and energy needed for the next duty. Charger nameplate totals are not the same as simultaneous depot import. Twenty vehicles each needing 60 kWh at the battery require 1,200 kWh in total. At 90% grid-to-battery efficiency, the grid must supply about 1,333.33 kWh before other depot uses.

Across an eight-hour shared window, the simple average charging demand is 166.67 kW. This is only an aggregate screen. If some vehicles arrive late or depart early, a feasible total-energy calculation can still leave a particular vehicle unready. Build a vehicle-by-time schedule and include charger availability, charge acceptance, operational reserve and a failed charger.

• ARENA: ANC last-mile delivery lessons, report 3

Suppose the depot has a 300 kW import cap and a constant 80 kW background load. Charging headroom is 220 kW. Eight hours at that ceiling allows 1,760 kWh from the grid, more than the 1,333.33 kWh fleet requirement. Four hours allows only 880 kWh, leaving a 453.33 kWh grid-energy gap. Minimum aggregate charging time is about 6.06 hours under the stated constant conditions.

Buying more chargers does not remove the upstream cap. Options include extending dwell, changing dispatch schedules, reducing other coincident load, using a properly sized charged battery or obtaining an approved capacity increase. Recalculate the complete schedule for each option. A controller needs a safe fallback when communications or load measurements fail; its existence does not establish permission to exceed the connection limit.

Useful heat and electrical input have different boundaries. For a synthetic process requiring 1.2 MW of useful heat, a heat pump with coefficient of performance 3 needs 0.4 MW electrical input at that operating condition. A resistance system with 95% useful-heat efficiency needs approximately 1.263 MW. A heat pump transfers environmental or recovered heat as well as consuming electricity; COP greater than one does not violate energy conservation.

A valid comparison requires supply temperature, temperature lift, heat source, output profile and integration with the process. COP changes with conditions and is not a guaranteed constant. Product quality, startup, turndown, maintenance and backup can constrain an otherwise efficient option. Do not assume one heat-pump example applies to every high-temperature process.

• Australian Government: process heat and steam

Add the proposed electrical demand to a time-resolved site profile. A constant 0.9 MW existing load plus the example 0.4 MW heat pump gives 1.3 MW, exceeding a 1.1 MW import cap by 0.2 MW. The electrical energy saving relative to another technology does not itself prove spare network capacity. Check transformers, connection terms and coincident operating conditions with the appropriate specialists.

For a simple useful-heat cost comparison, electricity at A$120/MWh and COP 3 contributes A$40/MWh of heat. Gas at A$12/GJ in an 80% efficient heater contributes A$54/MWh of useful heat because 3.6/0.8 × 12 = 54. These are energy-only contributions, not lifecycle costs. Add capital, network charges, maintenance, downtime, financing and relevant policy or contractual effects before making an investment recommendation.

A site with solar, a battery and a grid connection is not automatically capable of intentional island operation. Identify the loads that must survive an upstream outage and the source that establishes voltage and frequency when separated. Protection, earthing, controls, switching, restart and reconnection must form a coordinated, approved design. This lesson describes the decision boundary, not how to operate switching equipment.

For an energy-only screen, 80 kW of critical load for six hours requires 480 kWh delivered. A battery with 360 kWh available at the load boundary lasts 4.5 hours at that constant demand, leaving 120 kWh unmet. Solar output may improve a particular daytime case, but an annual yield or installed PV rating is not firm night-time outage energy. Separately check power capability, starting loads, losses and reserve margins.

• ARENA: lessons from Australian fringe-of-grid projects

Use one interval table for the underlying service, facility demand, flexible demand, local generation, storage and grid import. Identify contractual import/export limits, planned outages and project stages. Apply the same time zone, units and sign convention throughout. Then test a late vehicle return, an adverse heat-pump condition, unavailable generation or a longer outage.

Your recommendation should state the residual service gap, options, dependencies and decision gates. For an EV depot, the key result may be vehicles ready by departure; for a factory, useful heat at the required temperature; for a microgrid, critical load served through a defined outage. A cheap MWh is not a sufficient result if the actual service fails. Connect this topic to flexibility, storage, connection studies and procurement.

AEMO's 2026 ESOO Step Change scenario forecasts NEM data-centre electricity use rising from approximately 5 TWh in 2025–26 to 34 TWh in 2035–36. These are dated forecasts, not measured future outcomes or a national capacity register. Keep the NEM boundary distinct from Western Australia's WEM and the rest of Australia. Forecast scenarios, project connection enquiries and realised metered consumption answer different questions.

In a synthetic pipeline, 200 MW announced, 100 MW contracted and 40 MW energised must not be added to make 340 MW if they describe overlapping stages of the same projects. Keep a project identifier, milestone, expected utilisation and commissioning profile. A risk-weighted forecast may support a portfolio scenario, but it does not establish a particular site's right to connect or firm supply.

• AEMO: 2026 ESOO, p. 38 — Step Change data-centre outlook
• IEA: Key Questions on Energy and AI · April 2026

An AI training job may have some scheduling flexibility; an inference service may face latency and availability commitments. Neither label by itself establishes interruptibility. Ask which workload can move, how much power changes, how quickly, how long it can pause and what recovery load follows. A theoretical computing optimisation is not a bankable demand-response commitment until the service and controls are verified.

Higher equipment power density also changes the cooling design question. PUE measures total facility energy relative to IT energy; it does not measure water consumption, useful computation or resilience. Compare cooling options using temperature conditions, energy, water, maintenance and site constraints. A lower PUE can coexist with greater absolute electricity use as IT activity grows. Do not assert a universal water or energy advantage without a defined system boundary.

• IEA: Key Questions on Energy and AI · April 2026

A large load can matter through the speed, magnitude and coincidence of its changes. In a synthetic event, a 200 MW facility reducing import to 80 MW over 30 seconds changes demand by 120 MW, or an average 4 MW/s. That arithmetic describes the event; it does not establish acceptability. The response of controls, UPS, protection and other connected equipment requires appropriate network studies.

Record planned ramps, restart behaviour, disturbance response and credible common-mode events. A standby generator or UPS may support a site's service continuity without being permitted or configured for routine grid support. Separate the network connection envelope, internal resilience design and any proposed market-service participation. No calculation in this course authorises switching, protection changes or electrical work.

A campus consuming 100 MWh in each of two intervals needs 200 MWh in total. A renewable project producing 200 MWh in the first interval and zero in the second matches annual-style totals over this toy window, but leaves 100 MWh of time-specific shortfall and 100 MWh of surplus. A financial contract, an environmental attribute and a physically available energy supply are not interchangeable.

Compare commissioning dates, facility load shape, delivery location, contract volume, residual purchases and settlement terms. In a synthetic fixed-volume hedge, delaying campus demand from 100 MW to 40 MW leaves 60 MW of excess contracted volume before other positions. Decide who bears that risk, how it is managed and which milestones permit an increase in commitments. Use applicable scheme and contract rules before making a public renewable-energy claim.