The mining contractors had order books and lockectd-in revenues. For some investors, it felt safer owning companies that provided mining equipment or services, than owning the miners themselves.

Then, mining capital expenditure stopped growing – but that was enough to tip the scales.

The large miners, for the most part, survived. Many of the companies providing the picks and shovels for the mining boom did not.
 

AI capital may have the same problem

I wrote in this newsletter in August about the difference between a transformative technology and a profitable investment in it. The follow-on question may be the more practical one - where does the money actually go?

Moody’s has estimated global AI infrastructure spending above US$3 trillion through to 2030. [2] That capital cannot stay inside the technology sector. It has to buy land, power, steel, cooling, water rights, fibre and construction labour, and all of it has to be financed.

Every one of those is somebody else’s industry and company, and several of them are listed here in Australia.

A theme does not spread through a market because it is fashionable. It spreads because the capital behind it has to land somewhere physical, and the businesses it lands on get reclassified long after their earnings have already changed.
 

Risks of picks and shovels approach

The appeal of the ‘picks and shovels’ approach to AI investing is intuitive. Rather than guess which AI company wins, gain exposure to suppliers of products and services used in AI infrastructure. 

Webull believes this approach may work well for some businesses in this AI investment cycle. But the mining experience suggests caution.

Suppliers to a construction boom are not leveraged to how much AI infrastructure exists. They are leveraged to how fast more of it is being built.

Those are different exposures, and the difference tends to become visible only at the top. A data centre that has been built continues to earn for its owner; but the contractor who built it needs the next contract.

So, second-order businesses may not require a bust in AI infrastructure to be repriced lower by the market. A plateau in AI investment could be enough. Order books get thinner, and the valuation multiple attached to an AI services business might compress well before anyone declares the theme over.

That is not a prediction about timing. It is a reason to carefully consider which of those two positions you may hold 
 

Listed property is changing shape

Australian Real Estate Investment Trusts (A-REITs) is one area in which AI-related infrastructure activity may be reflected on ASX.

For a generation, ‘property trust’ meant shopping centres, office towers and industrial sheds. It was a yield story, bought for income by people who wanted less volatility rather than more.

Parts of that sector are now AI infrastructure businesses. The land banks, grid connections and planning approvals that took years to assemble are precisely what a data centre developer cannot obtain quickly, and several trusts appear to have redirected much of their development pipelines accordingly.

The classifications have not caught up. On one widely used measure of the Australian listed property market, data centres account for well under 1% of the sector, while industrial property accounts for almost 40%. [3] 

Both figures are accurate, but neither of them tells an investor how much of that industrial exposure is really a bet on computing demand.

And the mining distinction applies here too. A completed asset on a long lease to a profitable tenant is one kind of investment. A speculative development pre-committed to a tenant whose own revenue is still a forecast is another – but they can be easily misunderstood or misclassified by investors.
 

Diversified products, concentrated outcomes

The exchange-traded fund (ETF) market carries the same issue in a different wrapper. Australian ETFs held roughly $382 billion in assets under management across over 500 products by around September this year. [4]

But broad index products inherit whatever concentration the index already has, and the largest 10 companies on our own benchmark (the S&P/ASX 200 Index) account for close to half its value. [5]

In Webull’s opinion, thematic ETFs, which have drawn $1.3 billion of net flows in the first five months of 2026 alone, [6] do what they promise, but may overlap heavily [thematic ETFs typically invest in companies exposed to a chosen investment theme or ‘megatrend’.]

An investor holding a broad market fund, a global technology fund and an AI-specific fund may feel they hold three layers of diversification, when at the level of the underlying holdings they may have bought many of the same companies three times.
 

Two questions worth asking

None of this argues for avoiding AI-related investments, or for piling in. It argues for knowing which part of the boom you own.

• Am I buying the asset or the buildout? Owning infrastructure and supplying its construction are different businesses. Their financial outcomes may differ across stages of the infrastructure investment cycle 
• What would have to go wrong at the same time? Genuine diversification means holdings that fail for different reasons. If a construction slowdown would hit an index fund, a property trust and a thematic fund at once, that may be one position in three wrappers.
   

Conclusion

Australia’s mining boom was real, and the underlying thesis was correct. Australia’s iron ore and gas capacity is still here, still producing and still continue to operate and produce commodities. 

What caught some investors out was that the capital flowing into the buildout associated with the boom was mistaken for capital that could survive a plateau in the speed and scale of mining investment.

AI may well prove the more transformative of the two trends, but the plumbing works the same way. 

So, the useful question may not be whether to buy the AI theme, but whether you already own more of it than you’re aware of.
 

From ASX

ASX Content On-Demand is great way to keep up  with the latest market insights and trends through ASX podcasts and webcasts.

-------------------------------------
Sources

[1] Australian mining capital expenditure peak, 2012-13: https://www.rba.gov.au/publications/bulletin/2014/dec/pdf/bu-1214-3.pdf

[2] Moody’s US$3 trillion AI infrastructure estimate: https://www.bloomberg.com/news/articles/2026-01-12/data-centers-will-need-3-trillion-through-2030-moody-s-says

[3] Australian listed property sub-industry weights, data centres versus industrial: https://www.msci.com/documents/10199/1d02dade-f239-f976-584c-3cc105036da2

[4] Australian ETF funds under management and product count: https://www.moneymanagement.com.au/etf-market-to-hit-400bn-in-2026/

[5] ASX 200 top-ten concentration versus the S&P 500: https://firetrail.com/asx-200-index-concentration-great-when-it-works-terrible-when-it-reverts/

[6] Thematic ETF net flows, 2026 year to date: https://www.moneymanagement.com.au/thematic-etfs-grow-fum-24-fold-in-10-years/