Wednesday 28 October 2026 · Vol. 1 · No. 005 Sign inSubscribe
Best Effort

The weekday briefing on Australian telecommunications, data centres and AI infrastructure.

Data centres

CBA lifts feasible data centre pipeline to 7GW, with $260bn of investment by mid-2030s

Commonwealth Bank economists now see 7GW of feasible new Australian data centre capacity, up from 6GW, and about $260 billion of spending over the next decade. They say the 23GW of announced projects overstates what will be built, and that power supply is the nearest hurdle.

Economists at Commonwealth Bank have increased their estimate of feasible new data centre capacity in Australia from 6 gigawatts to 7GW, pointing to demand from artificial intelligence and cloud computing. On the bank's numbers, that build-out could lift total rated capacity to about 9.2GW by the mid-2030s, more than four times the current level, with construction activity peaking in the 2028 and 2029 financial years.

CBA expects roughly $150 billion of investment by 2030, which would support about 4GW of extra rated capacity. Another $110 billion after 2030 would bring cumulative spending to about $260 billion by the time the central 7GW case is complete in the mid-2030s. Specialised IT equipment, rather than buildings and structures, is expected to make up about 70 per cent of construction costs.

The feasible figure sits well below the headline numbers in project announcements. About 23GW has been publicly announced, yet most of it is a long way from construction. Some 3.6GW is being built or expanded and another 2GW has approval or is in pre-construction, which together make up 24 per cent of announced capacity. The other 76 per cent is still at the planning stage or earlier.

The Australian Energy Market Operator reported 225 projects seeking almost 67GW of connection capacity as of June 2026. CBA said that total probably exaggerates what will be delivered, because applications can overlap and developers may be chasing the same customers. More than a third of connection requests lodged in the previous year were withdrawn, and the median chance of a project advancing from its current stage to completion was put at 28 per cent. CBA Associate Economist Lucinda Jerogin said the "high degree of attrition and duplication in AEMO’s connection data demonstrates even meaningful progress through the connection process does not guarantee delivery".

CBA's central case is broadly in line with AEMO's electricity demand forecasts, which imply 5.5GW to 6.5GW of additional rated capacity by the 2036 financial year. The analysis also factors in possible demand linked to OpenAI and Anthropic, assuming in the central scenario that Australia captures 10 per cent of their estimated global compute demand. CBA described these as scenario assumptions rather than disclosed commitments. In an upside case where Australia becomes a major site for AI model training and large-scale inference, additional capacity could reach between 9.2GW and 11GW.

The report identifies energy infrastructure as the most immediate constraint. AEMO estimates data centres consumed 5 terawatt hours in 2025–26, equal to 2.8 per cent of electricity supplied through the National Electricity Market, and forecasts about 15TWh in 2029–30 and 34TWh by 2035–36. Community opposition to large developments has grown, and uncertainty over copyright law could affect whether large-scale AI training is hosted in Australia. Jerogin summed up the outlook by saying "Risks are two sided".