Beyond the gigawatts: what determines the return on Asia Pacific’s AI infrastructure
The return on compute depends less on how much gets built than on how well the rest of the economy can use it
Asia Pacific’s data-centre build-out is still accelerating. The region’s development pipeline reached a record 26.5GW in H1 2026, after another 7.1GW was added in just six months. Growth in capacity has, understandably, become the most visible measure of progress in the regional AI race.
But compute capacity is an input, not an outcome. What matters economically is how widely and effectively that capacity is put to productive use – and that depends less on the size of the build-out than on connectivity, access, adoption and the wider policy environment. Two economies could host similar data-centre capacity and get very different returns from it.
What the headline numbers miss
Governments tend to measure what is visible: capital committed, megawatts commissioned, construction jobs created. These track the investment cycle, not the return on it. Data centres are capital-intensive rather than labour-intensive, so their direct employment footprint is modest once construction ends. The lasting payoff, if it comes, shows up elsewhere – in the logistics firm, hospital or bank that uses AI and cloud services to work more productively. Those gains are slower, more diffuse and much harder to count.
The mismatch matters because the inputs are not free. Forecasts from Oxford Economics Australia and the Australian Energy Market Operator suggest data-centre electricity consumption in Australia could rise from 5.2 TWh in FY26 to 34.3 TWh by FY36 under the central scenario. In New South Wales and Victoria, data centres could account for around 19% of grid-supplied electricity by FY36. Not every proposed project will go ahead, but land, water, construction capacity and capital face similar claims. When compute could be drawing close to a fifth of the grid in Australia’s two most populous states, what the wider economy gets back is not an academic question.
Where the returns are made
Our recent research on other parts of the infrastructure system shows what makes the difference. Our work for Telstra on its A$1.6bn Aura intercity fibre network – linking major cities, data centres, cloud regions and subsea cable landing points – found that the benefits extend well beyond the network itself, as faster and more reliable connectivity supports greater use of cloud, AI and automation across businesses and sectors. Compute creates capacity; networks determine who can reach it.
Our research on low-Earth-orbit (LEO) satellite broadband makes the point more starkly. LEO could bring high-quality internet access to millions of people in unserved and underserved areas, and under the scenarios we modelled, wider adoption could raise global GDP by between US$32bn and US$863bn by 2035. That is a more than 25-fold range, and what drives it depends heavily on the conditions around it: competitiveness, affordability and adoption.
Policy makes the difference
How effectively digital infrastructure is used depends on a wider set of conditions including data governance, competition, skills, energy policy, digital trade and AI regulation – themes that we explored throughout this series. The first article in the series argued that the question is not whether individual technology policies are well designed, but whether together they allow technology to deliver its full economic potential. Digital infrastructure is where that question becomes most concrete – and most expensive.
These policies interact, sometimes in ways that cut against the instinct to simply keep on building. More open cross-border data flows, for example, might reduce the domestic capacity a country needs without reducing the economic gains available from AI. Energy policy decides how compute’s claim on the grid is balanced against other users. Competition and affordability shape whether smaller firms can access AI services at all; skills determine whether they can use them well.
The right mix will also differ across the region. Singapore’s geography and constraints pose a different challenge to Australia’s, and extending access across an archipelago such as Indonesia raises another set of questions again.
A broader measure of success
For governments competing for AI investment, success should not be measured by how many data centres are built but what that investment enables the wider economy to do. For investors, the same question translates into demand: whether businesses can access, afford, and productively use the capacity being created.
Asia Pacific’s AI boom is being measured in gigawatts. Its economic return will be measured in productivity – and that depends on what happens outside the data centre, not just inside it.
Explore more insights from our Economics in Tech Policy series, examining the economic implications of technology policy across APAC.
