Artificial intelligence and cloud computing are rapidly transforming how we live and work, from answering questions and creating content to powering the digital services around us. But behind every AI prompt, streamed video and cloud backup sits a network of data centres: large, energy-intensive facilities filled with thousands of high-performance computers.

Australia’s data centres are on the brink of a major expansion. Under the Australian Energy Market Operator’s central forecast, data-centre electricity consumption rises roughly sevenfold, from about 4.7 terawatt-hours (TWh) in 2025–26 to nearly 34 TWh by 2050. By mid-century, that approaches one-tenth of underlying electricity demand in the national grid. 

This growth presents an important question: will powering the digital economy place unnecessary costs on other electricity users, or can smarter operation substantially reduce those pressures?

A new submission to the Senate inquiry into artificial intelligence and data centres, led by Monash Energy Institute’s Dr Changlong Wang, finds that even modest flexibility in when some computing occurs could save Australia’s power system billions of dollars and reduce wholesale electricity-price pressures on households and businesses. 

The A$18.5 billion risk of inflexible growth

AEMO’s data-centre forecast treats their electricity demand as non-flexible. In other words, the power system is planned around that demand rather than assuming data centres can adjust their electricity use when the grid is under pressure. 

Using Monash University’s MUREIL-ISP national electricity-market model, aligned with AEMO’s 2026 Integrated System Plan, we compared a future in which data-centre demand grows as forecast and remains inflexible with one in which data-centre demand stays at its 2026 level. 

The results are substantial.

• A$18.5 billion in additional power-system costs. On a present-value basis from 2026 to 2050, inflexible data-centre growth adds A$18.5 billion to modelled power-system operation and expansion costs, equivalent to a 9.1% increase in total system cost. 
• Much more generation and storage is required. By 2050, the model builds an additional 7.9 GW of wind, 9.9 GW of solar and 3.9 GW of grid batteries, together with associated grid-security infrastructure. 
• Other electricity users bear much of the wholesale-price impact. By 2050, the modelling finds that inflexible data-centre growth increases the total annual wholesale electricity bill by about A$10.4 billion. Around 58% of that increase, or A$6.1 billion a year, falls on electricity users other than the data centres themselves. For a household using about 5 MWh a year, that wholesale-price effect is equivalent to roughly A$155 a year if fully passed through, before retail margins and other bill components. 
• The effect is concentrated in the tightest periods. The modelled wholesale-price increase averages about A$103/MWh in winter and rises by around A$341/MWh during the most stressed hours of the year, when electricity supply is tight and prices are highest.

These figures describe data centres supplied from the shared grid, consistent with AEMO’s forecast. A facility that arrives with firm generation or dependable self-supply would impose less additional demand on the system, particularly if that supply remains available during the scarce hours when costs are highest. 

Inflexible data-centre growth therefore costs consumers most in exactly the hours when the power system is already under the greatest pressure.

Not all computing needs to happen instantly

Not every computing workload has the same urgency.

Latency-critical services need to run when required. But some workloads, including AI model training, batch analytics, backups and offline inference, can potentially be shifted in time while still meeting their required completion deadlines. 

We tested what happens if 20% of IT computing work is delay-tolerant and can be rescheduled within a 24-hour completion window. The remaining 80% stays fixed in time.

Importantly, this is a modelling scenario, not a claim that every data centre can already shift 20% of its computing workload. The 24 hours is also a maximum completion window, not an assumption that every flexible job waits for a full day. Jobs may run immediately or move by only a few hours depending on system conditions, and every deferred job must still be completed. 

The results are striking:

• The power system saves about A$3.5 billion on the same present-value basis through to 2050. Increasing the flexible share to 30% raises the saving to about A$4.7 billion. 
• Consumers receive roughly 60% to two-thirds of the resulting wholesale bill relief across the flexible cases we tested. 
• Flexibility performs when the system needs it most. During the most stressed 5% of hours, flexible facilities delivered 96% of their promised flexible-demand reduction. 
• Long delays are not necessary to capture much of the value. In the central 20% flexibility case, a six-hour completion window already captures about 54% of the value available with a full week of leeway. 
• Data-centre operators benefit too. In the central and deepest flexibility cases, the flexible slice of computing pays an average wholesale electricity price of about A$24–38/MWh, compared with roughly A$140/MWh if those jobs were run when they arrived. 

The system benefits do not come from asking data centres to shut down. They come from shifting a limited share of delay-tolerant computing away from the small number of hours when electricity is scarce and expensive.

The modelling also suggests that timing matters. A flexibility commitment negotiated now, while new facilities are still being designed, is worth about 28% more than one phased in gradually as the data-centre fleet grows. 

Job-scheduling software is already standard technology in the industry. The bigger challenge is institutional: designing connection agreements, incentives and accountability arrangements so that flexibility can be counted on when the system needs it. 

As with any long-term electricity-system model, these results should be interpreted as scenarios rather than precise forecasts. The flexibility savings also assume well-coordinated scheduling, so they represent an upper bound on what real-world markets and contracts may capture. 

Key recommendations for Australia’s energy future

To capture this opportunity, the Monash Energy Institute recommends six policy steps.

1. Make flexibility part of the connection deal. Connection frameworks for large new loads should recognise and reward flexible capability and, where appropriate, expect facilities to declare what flexibility they can reliably provide. Flexibility does not necessarily need to be mandated by the system operator. 
2. Act in this decade. Governments should secure flexibility commitments while facilities are being designed rather than waiting until the fleet is already built. 
3. Keep the ask proportionate. Most of the value comes from moderate commitments: a minority of workload, relatively short completion windows and action during a small number of critical hours. Broad participation is more valuable than demanding extreme deferral. 
4. Pair 24/7 clean-energy matching with flexibility recognition. If governments introduce or encourage hourly clean-energy matching for large loads, workload flexibility should be recognised as a compliance pathway. 
5. Count what is already inside the building. Data centres contain substantial batteries, uninterruptible power supplies and standby generation. Connection standards should consider whether and how these assets can support the wider power system. 
6. Price the promise and the breach. If planners rely on promised flexibility when deciding how much infrastructure to build, that flexibility should be credited prudently, verified against actual delivery, and backed by consequences for non-performance. 

Australia does not have to choose between a thriving digital economy and an affordable, reliable power system.

By recognising flexibility when large data centres connect, and rewarding them for shifting delay-tolerant computing away from the tightest hours, Australia can reduce the infrastructure and wholesale-price pressures created by rapid data-centre growth.

The opportunity is to plan with data centres, not simply around them.

To read the full Senate inquiry submission, How to Grow the Digital Economy Without Sending the Bill to Households (September 2026), visit the Monash Energy Institute website.