A Senate inquiry launched on 13 May is probing the government AI data centre deals at the heart of Australia’s national AI strategy. The obvious questions (energy use, community impact, regulatory oversight) are the right ones to ask. But nobody in the terms of reference is asking where the numbers that justified those deals came from.

The figure driving Australia’s AI ambitions is $142 billion. That is the projected GDP uplift by 2030 that appears in policy submissions, ministerial press releases, and industry advocacy. It was prepared by consultant Shahar Merom, developed with a steering committee that included the Australian Computer Society, and funded by OpenAI.

The same OpenAI that is now the anchor customer for a $7 billion NEXTDC data centre in Western Sydney, welcomed by the minister whose former firm co-authored the blueprint.

What the report actually says

Australia’s AI Opportunities report breaks its $142 billion projection into three buckets: $112 billion from productivity gains through widespread adoption, $18 billion from building a domestic AI industry, and $11 billion from exporting AI-enabled products. It notes AI already contributes $21 billion annually and that realising the projection requires increased compute infrastructure, skills investment, and lighter-touch regulation.

OpenAI published the report on its own website as its “Economic Blueprint for Australia” in July 2025. The company was simultaneously lobbying Treasury, the Department of Prime Minister and Cabinet, and the Office of National Intelligence. Government briefing documents prepared before those meetings anticipated OpenAI would seek “Australian government investment in American infrastructure and the rolling back of our stricter approach to copyright and AI regulation.” Those same briefings noted that the wider benefits of AI “had yet to emerge.”

Six months later, the National AI Plan dropped the mandatory guardrails for high-risk AI that had been proposed by the previous science minister. OpenAI got that. It did not get the copyright changes or direct infrastructure co-investment, but it got the regulatory posture it wanted.

The Charlton layer

The firm that wrote the report is Mandala Partners, the offshoot of AlphaBeta. AlphaBeta was founded by Andrew Charlton, who was Australia’s Assistant Minister for Science, Technology and the Digital Economy when the report was published. Charlton later travelled to the US to meet OpenAI’s leadership and then welcomed the NEXTDC deal in a ministerial press release under his own name.

OpenAI also hired Kate Pounder, a former AlphaBeta partner and former Tech Council of Australia CEO, to lead its Australian policy work.

None of this is illegal. Consulting-to-government revolving doors are not new. Commissioned research is commissioned research. What matters is that the conflict of interest was not disclosed in the reporting of the $142 billion figure and was not flagged in the policy submissions that cited it. Founders, investors, and technology leaders making decisions based on these numbers deserve to know where they came from.

What the independent evidence says

The academic analysis from the University of Sydney puts numbers against the projection. Despite 40% of US adults now using generative AI, it accounts for less than 5% of work hours and has added less than 1% to labour productivity. General-purpose technologies like computers and the internet took decades to show up in productivity statistics. The report’s own case studies from Moderna and Canva contain “no data about improved organisational or individual performance,” relying on narrative claims.

The gap is not small. AI adding $142 billion annually to a roughly $2.7 trillion economy by 2030 requires a sevenfold increase in contribution over four years, against a measured productivity uplift that sits under 1%.

The Productivity Commission’s independent assessment (no OpenAI funding line) projects 4.3% labour productivity growth over a decade, roughly $116 billion cumulative over ten years, not per year. A real number, but a very different claim from the one in the pitch decks.

Why this matters if you’re building

The $142 billion figure is in the water. It shows up in investor rationale, board papers, government tender documents, and the pitch decks of anyone selling into the Australian market on an AI story.

Whether AI creates economic value is not the argument. It does. The argument is whether a specific projection, generated by a party with a direct commercial interest in a particular answer, should pass undisclosed through policy submissions and into capital allocation decisions. It should not.

The Senate inquiry that opened on 13 May is a late, partial correction. Its terms of reference cover energy and community impact, not the evidentiary basis of the economic case that got us here.

Treat vendor-funded projections the way you treat vendor-funded benchmark tests: a data point with a known bias, not a foundation for unexamined bets. The Productivity Commission’s work is a reasonable starting point. The OpenAI blueprint is a sales document.

If you’re making AI investment decisions for your business and want analysis that isn’t funded by someone selling infrastructure, talk to us.

Sources