Australia can secure the value of its AI future
As Prime Minister Anthony Albanese said in his speech at Sydney University last month, artificial intelligence is already a part of our daily lives and ‘no government can turn back the clock or press pause’. And nor should we.
Australia should instead pursue a strategy for selling AI-enabled products and services to the world while keeping returns onshore. This will require us to demand greater control over the repeatable benefits of AI and prepare for interruptions to computing power supply.
Australia starts from an enviable position in the transition to the agentic future: brilliant scientific development, innovators and entrepreneurs, world-class renewable energy resources, abundant land, trusted institutions and sophisticated industries. The challenge will be acting fast enough to capture enough of the value these assets create to control our own destiny.
Australia’s National Cabinet will meet this week to set AI policy, prioritising areas such as land, building, power, energy and data while largely ignoring who controls the output and what it costs. But these two factors will determine if, how and when sustainable value can be created.
With media attention set on data centres, the risk is that Australian AI policies treat computing capacity solely as infrastructure. We’ve seen this before. Australia’s policies on the internet, chip production and cloud computing all went the same way: Australia hosted the physical computers while the intellectual property, margin and tax base settled elsewhere. We bet on the wrong part of the value chain.
There are signs that history is repeating. Machinery and equipment investment, led by digital infrastructure hardware and software, jumped 16.3 percent in the first three months of 2026. In the same period, GDP grew just 0.3 percent, dragged down by net trade because most of that equipment was imported. The quarter’s biggest capital story barely touched the economy it was meant to drive.
AI outcomes have a price, calculated in terms of ‘tokens’, the small units into which AI models break down text, averaging about four characters each. More powerful models tend to have more expensive tokens. They also use more of them. Whoever prices the tokens has considerable influence over the economics of every industry running on AI.
Financial services company Goldman Sachs expects global token consumption to multiply 24 times by 2030, to 120 quadrillion a month. According to reporting by the Australian Financial Review, many organisations have reached (and surpassed) their annual AI budget allocations far earlier than expected. Models now show their working before answering, which costs far more tokens, and requests have spread from a pilot team to the whole company. The volume of requests is growing faster than prices. This means that while the same AI capability gets cheaper every year, bills keep rising anyway. In the boardrooms we sit in, discussions have shifted from exemplar use cases to ensuring acceptable return on investment.
Boards should look beyond average token costs. Analysis by Singapore-based AI aggregation company Ai.cc covering 2.4 billion enterprise application programming interface calls (requests between software applications) put the blended cost at US$6.07 (A$8.47) per million tokens, down 67 percent compared with the previous year. Companies routing tasks to the cheapest capable model paid a median US$2.31. Those sending everything to the frontier model paid US$18.40. Same intelligence, eight times the price. The difference is procurement discipline.
Token price decides what Australian industries can afford, and someone else sets them. Security agencies are already concerned this pushes enterprise toward open-weight models with near-frontier capability at a fraction of the cost. With many of these models originating in China, there is a risk for the nation – and the United States, our dominant technical alliance partner – that we remain a physical digital infrastructure winner that doesn’t control or create real value from the outputs of that infrastructure.
The government should consider three policies to help mitigate the economic and geopolitical challenges to AI supply.
First, build and maintain a domestic reserve of AI computing capacity, with guaranteed access for government agencies and industry if the supply we don’t control is interrupted.
Second, diversify where that capacity sits. Demand for computing power used to run AI models is projected to exceed demand for computing power used train frontier models by 2028. Increasingly, this computing power is run ‘at the edge’ – on local devices and computers rather than distant data centres. Yet more than 80 percent of our planned or active data-centre capacity is only in Canberra, Sydney and Melbourne – far from the mines, ports, communities or border posts the computing power will serve.
Third, require any vendor bidding for government or critical infrastructure work to disclose cost per outcome, not cost per GPU. A department that can’t tell whether it’s paying US$2.31 or US$18.40 per million tokens can’t tell whether it’s creating value for the taxpayer. Confidential deals with tech giants have never benefited Australia and won’t start doing so now.
The prime minister is right that digital infrastructure and its biggest beneficiaries need public support to operate. But the internet and cloud eras have shown that fast adoption in Australia while exporting the benefits offshore won’t lift productivity or living standards, and nothing on the National Cabinet’s agenda changes that. What should be addressed is the value AI creates here. Where the machines set is second order.
If Australia gets this right, it could export what runs on the hardware. Mining, agriculture, defence and health are industries where we already hold domain data and operating expertise no frontier model can buy. We have done it before: in 2020, Australia’s exports of mining equipment, technology and services were worth A$17 billion. But if we get this wrong, our ambitions for AI are moot and we cement a reputation as a well-regulated price taker.
About Author
