On November 8, the most famous man in computing, whose name you know and this column will spare, told his staff he had been unfaithful to his wife. According to, and as alleged by, American press reports, US officials have identified the woman as a former executive of a Singapore company said to have bought billions of dollars of his firm’s chips, and as an agent of Chinese intelligence. Beijing calls it a smear, no country has charged her with anything, and nothing made public shows he knowingly passed anything on.

His company’s shares slid regardless. Four days later he resigned as chief executive and left the board, effective immediately, after 33 years at the helm. The company gave no reason. Meanwhile, footage has surfaced of Australian data centre executives at the same industry events, proving only that the AI business is a very small party with a very long guest list. There is no suggestion of wrongdoing by any of them.

It is a soap opera, but it is also a stress test, and it forces a question Canberra has been too polite to ask: is Australia over-exposed to artificial intelligence?

We have no Nvidia of our own, no frontier model, no leading-edge chip plant, and no cloud giant. What we have, in abundance, is enthusiasm. We are exposed four ways: through our savings, our concrete, our power grid, and the fine print. On each count the honest answer is yes, by more than most Australians realise. And there is now an accelerant.

The savings

In an internal paper this September, Reserve Bank staff estimated that around 5.5 per cent of household financial wealth sits in AI stocks, about two-thirds of it through superannuation, and almost 90 per cent in foreign, mostly American, companies. A lasting 20 per cent fall in AI shares, they calculated, could cut household consumption by about 0.7 per cent, or around 2.5 per cent if the losses spread to the wider share market. The bank cautions that this may overstate the effect, which is the most reassuring thing a central bank can say while publishing it.

This is the quiet genius of compulsory super. Technology and communications firms now make up around 45 per cent of America’s S&P 500. A plumber in Penrith who ticked ‘balanced’ a decade ago now owns a large bet on a handful of Californian companies and their favourite chip supplier. Nobody asked him. That supplier is also the industry’s banker: by the Financial Times’ count earlier this year, it committed some US$90 billion to more than 145 AI companies in 16 months, and the man who personally negotiated the biggest of those deals has just left the building. AustralianSuper still held about US$1.5 billion of Nvidia at the end of June, even after cutting its stake by almost a third, which in this market counts as caution. The Reserve Bank’s October Financial Stability Review noted that exposure to the AI boom keeps growing while risk premia stay low. Investors are being paid very little to carry a great deal, and are delighted about it.

The concrete

The second exposure comes in grey. Commonwealth Bank economists reckon the data centre build could attract around $150 billion by 2030 and $260 billion by the mid-2030s; the quantity surveyors Rider Levett Bucknall think it could rival the mining construction surge of the 2010s.

Australians will recognise the tune: last time we dug things up, shipped them to one enormous customer, and acted surprised when its mood set the price. This time the customers are a few American technology giants. Amazon has pledged $20 billion; OpenAI has signed a memorandum for a $7 billion campus in Western Sydney. Lovely. But a building full of someone else’s chips, running someone else’s models, earns us rent, not royalties, and CBA expects the whole surge to add only about 0.2 percentage points to growth in each of 2026 and 2027. We have become the landlord of the future, and landlords are rarely the ones inventing anything.

The grid

The third exposure plugs into the wall. The energy market operator expects data centres to go from about 3 per cent of the eastern grid’s consumption to around 13 per cent by 2035-36, just as a large slice of coal and gas generation heads for retirement. Data centre projects seeking connection add up to almost 67 gigawatts, which is either boundless confidence or a lot of people queueing for a lottery. In June every energy minister except Queensland’s agreed that data centres should fully offset their demand with new renewable supply. Sensible, if it holds. But communiqués have a shorter half-life than turbines, and if the promise slips the bill will be posted to Australian households.

The fine print

The fourth exposure is the one we understand least: How many of America’s best chips end up in China anyway? The research group Epoch AI estimates that smuggled chips made up roughly a third of China’s AI computing power by the end of 2025, though it acknowledges its range is wide. It also finds trade data consistent with more than US$3 billion of chips reaching China via Malaysia, a neighbour that shares our region and our sales pitch. The Singapore company at the centre of the scandal is reported to have run data centres in Malaysia and Indonesia that appeared to serve Chinese customers remotely; it says Singapore’s own investigation ended with no adverse action, and Nvidia says it found no evidence of diversion. Washington restricted the hardware, so the pressure moved to rented computing next door.

The Deputy Prime Minister assures us Australia is ‘very alive to this’, but our foreign investment rules are built mainly around who owns a data centre, not who rents its capacity. We check the deeds carefully and wave the tenants through.

The accelerant

Now the code is learning to write itself. In an amended filing for its share market float, lodged with America’s securities regulator on November 2, the AI company Anthropic warned that its future models may ‘imminently’ meet its own definition of dramatically accelerated progress driven by AI: the threshold associated with recursive self-improvement, in which AI helps build the next, better AI. The company stresses that no current model has crossed the line and that it intends to keep humans meaningfully involved. It also concedes that AI-assisted development could outrun the safety measures meant to keep pace, and that it has not always recognised the size of a capability jump until after the fact.

Note who is saying this: not a protest group with placards, but a company preparing to sell shares to the public, in a document its lawyers have to stand behind. If the code starts improving itself, the value of owning it compounds with every generation. The value of housing it does not. A landlord’s rent rises with demand; a royalty rises with every improvement. We have chosen the rent.

So, over-exposed?

Yes, in the sense that matters: the bet is concentrated. The problem is not the size of the wager but its shape. We carry the downside, in super balances, power bills and grid risk, while the upside, in the chips, the models, and the intellectual property, belongs to someone else. In the mining boom we at least owned the ore. If the machines are about to start building their own successors, that gap will widen faster than any Senate inquiry can schedule a hearing.

None of this argues for turning the boom away. It argues for three modest fixes, each dull enough to have a chance. Super funds should tell members, in plain language, how much of their balance rides on a few AI names. The data centre offset rule should be written into law, not left in a communiqué. And anyone renting large blocks of Australian computing capacity should face the same scrutiny as anyone buying the building.

AI may well be the future, but a country that owns the concrete and not the code should be careful about calling it a sure thing.

The post We own the concrete, not the code appeared first on The Spectator Australia.

Leave a Reply

Your email address will not be published. Required fields are marked *