The £400 Billion Figure Was Never the Government's Own

What the Minister Actually Said

Kanishka Narayan, the AI minister, answered a parliamentary question and set the record straight in the driest possible terms. AI adoption and data investment, he said, could add between £55 billion and £140 billion to the UK economy by 2030, with a central estimate of £120 billion. Even that figure, he conceded, is a “hugely uncertain” estimate. The £400 billion, by contrast, “was drawn from a publicly available 2023 report commissioned by Google and produced by Public First,” a British consultancy now referred to as PF Global. [2] This is the first recorded evidence WIRED could find of the government placing the two numbers side by side, or acknowledging the Google commission at all. [2] The answer did not withdraw the larger figure. It did not declare it wrong. It simply let it stand in the record, a ghost that had already done its work.

How a Borrowed Number Became Policy

Government departments and agencies used the £400 billion figure for almost two years, across two administrations. Ministers have had the lower estimate since at least November 2025, nearly a year before the admission. Yet Skills England, the agency responsible for apprenticeships, was still publishing the old figure four months later. That lag is not a clerical error. It is what happens when a number enters the bloodstream of government before anyone asks where it came from. The £400 billion did not need to be true to be useful. It needed only to be repeated often enough that questioning it felt like questioning the technology itself. The lower figure — £120 billion — is still a large sum. It is simply a sum that someone inside government actually calculated, rather than one borrowed from a company with a commercial interest in the answer.

The Consultancy’s Rebuttal

PF Global, the consultancy that produced the Google-commissioned report, offered a partial explanation for the gap. Jonathan Dupont, its head of economics, said the disparity was partly due to what the firm sees as slow government adoption of AI. The technology moves fast, he argued, and the assumptions behind any model have to be updated as evidence arrives. The government’s numbers, he said, assume a much slower rate of adoption than seemed possible when the 2023 model was built. The firm also pointed to more top-down adoption and integration into civil service workflows as examples of how take-up could improve. Its central estimate for the amount of potential likely to be achieved by 2030, Dupont said, is now much closer to the government’s number, with growth ramping up in the next few years after that. This is a reasonable defence, and it contains a genuine insight: the gap between forecast and outcome is not only about the technology’s potential, but about how quickly an institution can absorb it. A slow bureaucracy can make a good technology look mediocre.

The Committee Chair’s Diagnosis

The £400 Billion Figure Was Never the Government's Own (Image 1)

Dame Chi Onwurah, the Labour MP who chairs the Commons Science, Innovation and Technology Committee, named the problem directly. [1] The government is right to promote AI’s potential to improve public services, she said. But these reports highlight one of the central barriers to delivering digital transformation that her committee has identified: hype. [1] Economic projects inevitably rely on assumptions, she noted, but overstating the benefits of technology risks undermining the case for change. The government needs to be realistic about what AI can deliver and transparent with the public about potential savings. Her word — hype — is the precise one. Hype is not a lie. It is a number repeated without its provenance, a forecast stripped of its caveats, a report paid for by an interested party and cited as if it were independent. Hype is what fills the space where evidence should be, and it is corrosive precisely because it sounds like confidence.

A Peer’s Long

Wait for an Answer Lord Holmes of Richmond, the Conservative peer, asked ministers in January 2025 about the origins of the £400 billion claim. He did not receive a response. His verdict, delivered to WIRED, is measured but pointed. [2] It is unfortunate, he said, that the figure used repeatedly by the government, and then unsurprisingly by multiple other organisations, has now been significantly and rightly adjusted. It would have been better for the government to clarify this much sooner and more directly. He added that robust data is critical, not least because billions of pounds of public money, and how the government chooses to deploy it, depend on it. His phrase — “and then, unsurprisingly, by multiple other organizations” — captures the mechanism exactly: a government figure becomes an industry figure becomes a media figure becomes a public assumption. A government figure becomes an industry figure becomes a media figure becomes a public assumption. Each repetition adds authority and subtracts scrutiny.

The Research Landscape: What We Know and What We Assume

The dispute over two numbers is really a dispute over a method. Sumedha Deshmukh of the Ada Lovelace Institute described the pattern in plain terms. The government’s revised estimates are welcome, she said, but it is concerning that they come after almost two years of ministers citing a figure from research that was paid for by Google. Figures about AI that look precise often have assumptions and incentives built into them, and industry often funds or shapes the research behind them. Her prescription is procedural rather than dramatic. The government should test the numbers it is given before it builds policy on them, fund independent research, and measure whether promised gains are delivered, she said. It should always start by asking simple questions — where did the number come from, what does it assume, and who actually benefits from it. These are not technical questions. They are the questions any careful reader would ask of a claim in a newspaper, applied to a claim in a budget.

Where the Real Gain Lives

Strip away the borrowed figure and something concrete remains. The government’s own central estimate — £120 billion by 2030 — is still a substantial projection, and it rests on a mechanism that can be observed rather than asserted. AI adoption in civil service workflows, data investment, and the integration of tools into existing processes are measurable activities. They produce measurable outputs: hours saved on administrative tasks, faster processing of claims, more consistent analysis of large datasets. The gain is not a headline number. It is the accumulated effect of thousands of small improvements, each of which can be audited. That is a less exciting story than £400 billion, and it is the only story that survives contact with evidence. The Ada Lovelace Institute’s advice to test the numbers before building policy on them is the spine of any honest account of AI’s economic value. A number that cannot be checked is not a forecast. It is a slogan.

The Reversal at the End

The £400 Billion Figure Was Never the Government's Own (Image 2)

Here is the data point that reverses the frame. The £120 billion central estimate is not a disappointment. It arrives with a caveat the £400 billion never had: it was calculated by people accountable to the public. The lower number is more useful than the higher one precisely because it is smaller and openly acknowledged to be uncertain. A government that cites a modest, auditable figure can be held to it. A government that cites a borrowed, inflated figure can only be embarrassed by it. The admission last week was not a confession of failure. It was the first moment in two years when the public was given a number it could actually test. That is what a genuine gain looks like in policy: not the loudest claim, but the one that survives being questioned. The £400 billion was never the government’s own. The £120 billion is, and that is why it matters.