I think AI changes everything. Not most things. Everything.

Don’t brush that off as mere enthusiasm. It’s a conclusion that comes from a long habit of trying to take exponential curves seriously. Moore’s Law was simply the training. It ran for decades, and at every point along it there were sensible people extrapolating linearly and being wrong, because a doubling is not an increment and human institutions are built for increments. The scaling laws are the same shape. More compute, more data, more parameters, and the capability curve keeps going where a linear reading says it shouldn’t.

The usual mistake with this kind of argument is being right about direction and early about timing. Believe me, I’m familiar with it. You see the curve, you believe it, and you sit there for eight years while the world takes its time.

That isn’t the failure available here. Capital expenditure on AI infrastructure is at record levels, which is money already committed rather than opinions being expressed. AI is on the cabinet agenda of every government in the developed world. The build is happening now, and it is happening on a schedule that nothing physical was designed to meet.

Everything below is my own reasoning about my own money, and is not a proposition for anybody else’s.

H2: Exponential capital, linear steel

Here is the general form of the problem, and I’ve spent two posts on specific instances of it without naming the shape.

Compute compounds. Chips get faster, models get more capable, capital commits at software speed. Steel, however, does not compound. A transmission line takes a decade. A turbine has an order book. A grid connection has a queue, an approval has a process, and none of those things get twice as fast every two years no matter how much money is standing behind them.

That gap is the whole story. In the first post it showed up as speed to power; the reason a Singapore developer put $31.9 billion of proposed data centre on a cattle feedlot next door to two gas-fired power stations. Because the shortfall was not fuel but years. In the second it showed up as a choice about what a state exports, because once compute starts moving toward energy, an energy producer can sell the molecule or sell what the molecule makes.

Exponential demand meets linear supply, and the adjustment happens wherever the system is softest. It’s the same collision seen from different angles.

The institutions are visibly behind. Unfortunately, that’s less of a criticism and more of a recognition. AEMO now assumes that a meaningful share of the projects in its connection queue will never be energised. National Cabinet is drafting standards on data centre energy and water with legislation planned for next year, while Queensland argues about them, and a thirty-billion-dollar application sits in front of a regional council. Everyone is working at the pace of process on a problem moving at the pace of capital.

H2: Which is why Australia says yes

The policy argument currently taking place is a fairness argument. Should hyperscalers get to draw on a grid that households paid for? Should existing users subsidise the connection? And underneath all of this is the bigger question: is any of this a good deal for the country?

They’re reasonable questions, and I don’t think they’re going to decide the outcome.

Australia has run this argument before, and the pattern is easy to spot. In the 1960s, the Pilbara was a foreign-capital question: enormous resource, someone else’s money, a country arguing about the terms. The money came, the industry got built, and sixty years of wealth flowed from it to the state, the Commonwealth, and the region. Not everyone had a seat when those terms were set, and Traditional Owners waited three decades for native title to be recognised. The Bowen Basin ran the same course. So did Curtis Island, more recently and more contentiously. The argument isn’t won, so to speak. It gets overtaken.

Zerra DC is that pattern arriving again with a different substrate. The common factors are all there: foreign capital, Australian energy, a regional site, and a local debate about water and power. I’d expect the standards to be written, softened where they would block a build, and then built around. Nobody needs to fold for that to happen. A state watching thermal generation retire across the national market over the next decade, while an inbound investor offers to underwrite new generation, has a very predictable set of incentives.

That’s a prediction rather than a preference, and I’d sooner be accurate than righteous about it.

H2: What Australia captured last time

Now the part that matters for an angel, because none of the above is reachable. You cannot invest in a $31.9 billion data centre and neither can I.

Let’s look back to the Pilbara and ask what Australia ended up owning. Not the ore bodies, since they’re mostly held by two very large companies. Not the downstream processing, which we lost on freight, and which I’ve argued we won’t get back. What we ended up owning was the technology layer. Mining software, remote operations centres, autonomous haulage, geotechnical monitoring, drill-and-blast optimisation, equipment health, water management, and so on. A genuine export industry, built by companies most Australians have never heard of.

It happened for an unglamorous reason. The hard physical problems were here first. If you are operating in the Pilbara, you cannot buy the solution from somewhere else because nowhere else has your problem yet. Someone local solves it, badly at first, and by the time the rest of the world has the same problem there is an Australian company ten years into solving it.

Being early to the problem is worth more than being early to the market. That’s the mechanism, and it has nothing to do with mining.

H2: The same layer, a different substrate

When you put a hyperscale AI campus in a gas field 250 kilometres from a city, you generate a set of problems nobody has good answers for. This is the ground the data centre energy startups will grow out of, and it looks like this:

• Microgrid orchestration. Solar, wind, batteries, gas firming, grid supply and a load that must not blink. Real-time optimisation across all of it.
• Cooling without water. Rack densities climb hard and Australia is dry. Every cooling approach that assumes abundant water is a non-starter in Toowoomba’s catchment.
• Speed to power. Connection studies, congestion modelling, grid engineering, approvals. Anything that takes months off an energisation date is worth a great deal to someone.
• Workload scheduling against a price. If some compute israble, a data centre stops being a passive load and becomes a demand-response asset. Somebody has to write that.
• Gas optimisation. Field automation, compression, methane detection, emissions monitoring, gas-to-power integration.
• Power electronics. Transformers, converters, UPS, high-voltage DC, switchgear, condition monitoring at densities the equipment wasn’t designed for.
• Provenance and attestation. Once workloads move between jurisdictions, proving where computation happened and what powered it becomes a product.

Almost none of these companies will call themselves AI companies. They will look like an industrial software business, a cooling business, even a grid engineering business. Which means the people scanning for the letters A and I will miss them, and the people who understand energy systems will find them first.

Brisbane has those people. This city is full of engineers who have spent twenty years on remote operations, industrial automation, power systems, gas fields and water management, because that is what the Queensland economy has been doing. That skill set was assembled for coal and gas, and it transfers almost intact.

We’ve already backed one. MooCoo is an investor in Enaxiom, which approaches data centre cooling in a different way so that ultra-pure water comes out as a by-product rather than going in as an input. I’m not putting that forward as a recommendation to anyone, and it may or may not work. I’m putting it forward as evidence that the category is not hypothetical, and the deals are already crossing our desk.

H2: Where the chain could break

This is an argument in links, and any of them can fail.

On-device inference could pull enough computation out of centralised facilities to change the shape of demand. Data residency rules could harden until almost nothing is geographically contestable. Cheap long-duration storage could make renewable electricity effectively firm everywhere, which would remove the advantage an energy region has. Or AI could disappoint commercially and the capex could stop.

I don’t think any of those will happen, but I’ve held the first three posts of this sequence together with reasoning rather than evidence, and reasoning compounds errors as efficiently as it compounds insight.

H2: Living in the future doesn’t tell you which company

Here’s where I have to be careful with myself, because this is precisely the point at which a good macro argument turns into a bad investment decision.

I’ve talked before about why optimism is the right seat for an angel. If you don’t believe the future is bigger than the present, you shouldn’t be funding it. Living in the future is the job. Working out what happens next and standing there early is the job.

And none of it improves my ability to pick the winner.

Ten data centre energy startups will be founded in Brisbane over the next three years. I’d guess one of them returns everything and most return nothing, which is the base rate for this asset class. Being right about the sector doesn’t move it. Conviction about a future and conviction about a company are different things, and the second one is almost always unearned.

So the thesis does one job. It tells me where to look, and it stocks the pond. Everything after that is the same discipline it always was: small cheques, a lot of them, and a hard question about whether I can articulate why each one fails.

What I’m confident about is narrower and more checkable than it sounds. The capital is committed, the physical system can’t move at its speed, and the people who will build the companies that close that gap are already in this city, currently employed by somebody else.

That last part is the bit worth being early on. Ask me in two years.

Richard Moore is co-founder of MooCoo Ventures, an angel syndicate that co-invests alongside Brisbane Angels, one of Australia’s most active angel groups. He has made over ninety personal angel investments since 2013.

moocoo.vc

DISCLAIMER: The information provided is for general informational and educational purposes only and should not be construed as financial, investment, or legal advice. Any decisions you make based on this information are your sole responsibility. You should consult with a qualified financial advisor or other professional before making any financial decisions. Past performance does not guarantee future results.