Data centre fault ride-through means a facility keeps drawing power from the grid through a short voltage disturbance, instead of switching its load to UPS batteries or generators. One facility disconnecting is a local event. Several large facilities in the same area disconnecting together, from the same fault, is a system security problem. Designers can fix most of it with UPS settings, protection settings and controlled reconnection.
The issue has moved from theory to rule-making because of scale. The data centre forecasting overview published with AEMO's 2026 Electricity Statement of Opportunities (ESOO) in August 2026 counted 225 data centre projects in development, with 67 GW of combined capacity, up from roughly 38 GW in the 2025 ESOO. Even if a large share never proceeds, the load that does connect will cluster around a few metropolitan transmission nodes. That clustering is what turns a design detail inside a building into a question for the whole NEM.
What Happens During a Fault
A transmission fault is cleared by protection in roughly 100 to 200 milliseconds on most of the network, longer for backup clearing or distribution faults. During that time, voltage at nearby busbars can sag well below normal. Data centres a long way from the fault still see a dip, just a shallower one.
The UPS does its job
A data centre is designed so that IT load never sees that dip. In a typical double-conversion UPS, the rectifier feeds the DC bus and the battery sits on it. When input voltage or frequency moves outside the UPS's input tolerance, the rectifier stops drawing from the grid and the battery carries the load. Line-interactive and other topologies do the same thing with a transfer switch.
From the IT hall, this is a success. From the grid, the facility's demand has just vanished in a few milliseconds.
The load does not come straight back
What happens next depends on the facility's control philosophy. Some UPS systems walk back onto the grid within seconds once voltage recovers. Many facilities go further: an undervoltage relay or the building management system starts the generators, transfers to them, and only retransfers to the grid after a stable period, often several minutes. Where that is the design, a 150 ms fault can remove a data centre's grid demand for five to fifteen minutes.
Simultaneous load loss and the frequency rise
Now multiply that across every data centre electrically close to the fault. If they share a similar UPS input window, they all drop together. The NEM then has more generation than demand:
- Frequency rises. A sudden loss of load acts like the mirror image of a generator trip. Lower contingency FCAS and generator governor response pull frequency back, but they are sized for credible contingencies the system expects.
- Voltage rises. Lightly loaded transmission lines generate more reactive power. Losing a large block of load near a node can push local voltages high just after the fault clears.
- The reconnection is a second event. When the facilities transfer back, the load returns. If they all come back at once, that is a step increase in demand the system has to absorb.
Historically, the NEM's largest credible contingencies have been generator or interconnector trips. A cluster of data centres reacting to one fault can produce a load loss of the same order, and AEMO cannot plan for it if it cannot see it coming.
International Lessons
The concern is not hypothetical. NERC's incident review of a July 2024 event in Northern Virginia describes a single 230 kV line fault after which about 1,500 MW of voltage-sensitive data centre load transferred to backup power, a loss of load the system operators had not anticipated. The findings translate directly to the NEM:
- The facilities behaved exactly as designed. Nothing failed.
- Their UPS and backup transfer logic responded to the same voltage dips, so they acted as one large load.
- The system operator had little visibility of the aggregate behaviour before the event.
- Reconnection went without major problems on that occasion, but NERC warned that future events could cause issues if load does not return in a controlled way.
We would point any developer to the NERC report itself rather than secondary summaries. The details matter, and the lesson for Australia is about the mechanism, not any single headline number.
What the NEM Is Proposing
The AEMC's "Improving the NEM access standards, Package 2" (ERC0394) draft rule of 12 March 2026 defines a large inverter-based load in the Rules for the first time, at 30 MW, where the working figure had been 5 MW, and puts ride-through and recovery of pre-fault demand at the centre of the new standards. The final determination was targeted for 29 October 2026 but had not been published as at 6 November 2026, and a separate AEMO rule request lodged on 21 July 2026 covers operational visibility of large loads. Our article on access standards for large inverter-based loads sets out the tiers, the status of each workstream and what to check in the final rule.
Design Levers
The good news is that ride-through is mostly a matter of settings and coordination, not exotic hardware. The levers we look at first:
UPS input tolerance and transfer thresholds
Most UPS platforms let the operator adjust the input voltage window and the time delay before transfer. Widening the window, or adding a short delay so a shallow, brief dip does not trigger transfer, keeps the rectifier drawing from the grid. The constraint is the UPS manufacturer's limits and the IT load's own tolerance, which the UPS still protects.
Rectifier control
Modern active front-end rectifiers can ride through a dip while drawing reduced current, then ramp back up. Their controls determine whether the facility recovers demand smoothly or with a step, and whether it draws or injects reactive current during the dip. That behaviour has to be specified and then shown in the models.
Protection and generator transfer logic
Intake undervoltage relays and building management system logic often decide whether a fault becomes a ten-minute outage from the grid's perspective. Settings should distinguish a cleared transmission fault from a genuine loss of supply. That usually means coordinating voltage and time settings with the network's protection clearing times, which is exactly what a protection study does.
Staged reconnection
Where a facility does disconnect, its return should be ramped or staggered across UPS modules and halls. A defined ramp rate is far easier for AEMO and the network to manage than a single block.
On-site BESS
A large battery at the intake can support voltage, absorb the transition and smooth reconnection. Our article on why data centres are turning to BESS covers the reliability case in more depth.
Studies Needed
A data centre connection now looks a lot like a generator connection from a modelling point of view. Expect the network and AEMO to ask for:
- An EMT model of the facility. A PSCAD model that represents UPS rectifiers, their control and protection, transformers, and any on-site BESS or generation. Inverter-based behaviour inside the first cycles of a fault cannot be captured in RMS tools alone.
- An RMS model for wider studies. A PSS/E representation for load flow and dynamic studies, validated against the EMT model. Our guide to PSS/E and PSCAD modelling explains how the two fit together.
- Fault ride-through studies. Faults at varying distances and clearing times, showing whether the facility stays connected, what current it draws during the dip and how fast demand recovers.
- Aggregate behaviour studies. The network may study your facility together with neighbouring large loads to check whether a cluster stays within credible contingency limits.
- Model validation at commissioning. Expect a process similar to the hold point and R2 validation generators already go through. The AEMO generator performance standards process is a reasonable preview of the discipline involved.
The common failure point is vendor data. UPS and rectifier manufacturers are not used to supplying validated grid models, so ask for them at procurement, not after the connection application is lodged.
Engineer Your Facility's Ride-Through
The cheapest time to get ride-through right is before the UPS and protection philosophy is locked in. Changing a threshold on paper costs almost nothing. Retrofitting after a failed model review costs months.
ElectraGlobe builds facility EMT and RMS models, runs fault ride-through and protection studies, and supports large-load connection applications through to the connection agreement. See our data centre grid connection service, or contact our team to talk through your facility.
Frequently Asked Questions
Why does it matter if a data centre disconnects during a fault?
One facility disconnecting is manageable. The risk is many large facilities near the same fault disconnecting together, which removes a large block of demand in milliseconds and pushes frequency and local voltage up. If they then reconnect together, the system sees a second disturbance.
Can UPS settings be changed to improve ride-through?
Often, yes. Many UPS platforms allow the input voltage window and transfer delay to be adjusted within the manufacturer's limits, and rectifier controls can be tuned for smoother recovery. Any change needs to be checked against IT load protection and confirmed in the facility model.
Will ride-through rules apply to existing data centres?
That depends on the final AEMC determination and its transitional arrangements, which had not been published as at 6 November 2026. Access standards usually apply to new connections and to existing connections when they are materially altered, such as a capacity expansion. Operators planning an upgrade should assume the new framework could apply.
What models does a data centre need for AEMO?
Expect a PSCAD EMT model of the facility, including UPS rectifier control and protection, and a validated PSS/E model for wider network studies. The exact requirements will follow the final access standards and AEMO's model guidelines, so confirm them with the network at the enquiry stage.