Somewhere across the state line from Memphis, in Southaven, Mississippi, there is a supercomputer that runs on a power plant that, on paper, is not a power plant.

The computer is Colossus 2, xAI’s cluster for training and running Grok. It needs an enormous amount of electricity, more than the local grid could deliver on the timeline xAI wanted. So rather than wait for the utility to build, xAI trucked in gas turbines (59 of them by the most recent count) and classified them as temporary, mobile equipment. That’s roughly double what it had publicly admitted to.  

I caught all of this on the All-In podcast during my morning walk, which is where a good many of my pieces start. There’s a lot to unpack, particularly with an understanding for the politics at play. Temporary and mobile turbines do not need the same federal permits a fixed power station does. Critics call it a fixed power station wearing a hi-vis vest and pretending to be a hire car. The Clean Air Act fight is now live, the community downwind is not happy, and in an intervention you do not see every day, the US Department of Justice has weighed in on xAI’s side on national-security grounds: Grok, the argument goes, is tied to US military capability, and switching off the data centre would hand ground to America’s adversaries.

Hold that story for a moment, because it is the newest instalment of a very old venture pattern, and the pattern is the point.

Everyone’s favourite trophy deal was built in the grey

The returns that make a fund’s decade are disproportionately made in the grey area. It’s the space where a business is operating ahead of the law, or against an entrenched incumbent, or on the wrong side of a rule that has not caught up yet. It’s uncomfortable, because we prefer our winners clean. But look at the roster.

YouTube, in its early years, was substantially a library of things other people owned. From clips of Saturday Night Live to music videos, and highlights from television. The infringement was not incidental to the growth; it was a good part of the engine. Google bought the problem along with the company and then did the expensive, grown-up thing: striking licensing deals and building the content-ID machinery that turned a copyright liability into a copyright business.

LinkedIn grew by reaching into your address book and inviting everyone in it, in your name, whether you had quite understood that or not. Others tried versions of the same move and choked on it, because it looked spammy and it was spammy. LinkedIn got sued. In the Perkins settlement, it paid US$13 million due to the way those reminder invitations used members’ names. Thirteen million dollars from a company that would list on the NYSE valued in the billions and later sell to Microsoft for US$26 billion. The fine was a blip, compared to the network it had bought with the tactic.

Uber is the interesting one, because unlike LinkedIn and YouTube, Uber started clean. The original product was black cars and licensed limousines with a premium service operating well inside the rules. The peer-to-peer move, ordinary people driving their own cars for money without a commercial licence, belonged to Lyft and Sidecar, and it was flatly against the licensing regime when they launched. Uber’s answer was UberX, in July 2012, matching the move it had not made first. Then Uber went hard, harder than anyone, flooding city councillors with tens of thousands of emails, mobilising riders as a political army, fighting the taxi lobby city by city. Walk through any major airport today and you will see the rideshare zone is full and busy. The taxi rank, where it still exists, is the quiet one. The lesson is not that the rule-breaker always wins. It is that the grey area is not always where a company begins. Sometimes it is where the company gets dragged, and then out-executes the one who opened the door.

Airbnb ran along a spectrum, starting literally on the floor with a blow-up mattress, and nobody seemed to mind. Offer a spare room, and it’s probably fine. Make it a whole apartment run as a de facto hotel, and the hotel industry starts screaming to anyone who will listen, with a fair point about safety rules and taxes it must follow and the apartment does not. From personal experience, I would far rather have a kitchen, a laundry, and a second room than a sterile hotel box. On long trips the apartment wins for me every time. The market, by and large, agreed, and the law bent toward the demand.

Two ways through the door, and they are not the same bet

Line those up and they split into two mechanisms.

Airbnb and Uber, in the end, won on legitimacy. Society tried the product, preferred it, and the law adapted to the preference. The customers voted with their dollars, and the regulation followed the vote. The empty taxi rank is what a settled social preference looks like once it has hardened into infrastructure.

YouTube and LinkedIn won on something closer to brute force. There was no groundswell of public affection deciding the copyright question or the spam question. Those were fought, absorbed, paid for, or simply outlasted. YouTube had Google’s balance sheet to convert the fight into licensing. LinkedIn had a network already built by the time the fine landed. The public did not ratify the tactic. The company survived it.

Two different bets, then. A legitimacy bet is a bet on where society’s preference will land. A brute-force bet is a bet on surviving the collision, on having the resources or the head start to still be standing when the incumbent finishes swinging.

The truth is that every situation could have gone the other way. If the hotel lobby had galvanised earlier, if the taxi licensing bodies had moved faster, if the record labels had picked a different fight at a different moment, several of these trophies would be tombstones instead. This is what I think of as sliding-doors risk. The outcome turned, in part, on how well the incumbent played a hand that the founder did not control. It is genuine risk. When one of these fails it can be nobody’s fault, in the sense that the founder did everything right and the door still swung shut.

So how do you accept a risk you cannot eliminate?

You cannot diligence sliding-doors risk out of existence. And you shouldn’t avoid the grey area, because the returns live there and the founder frequently has no choice but to operate in it. Instead, the question should be how to hold that risk with your eyes open. The answer, the thing I have come to think is the real intellectual work of angel investing, is that you need a view on the politics.

I’m talking less about political parties and more about political climate. Politics as in: where is society going to land on this, and when? Which way is the regulator leaning, and who appoints the regulator? Am I underwriting a legitimacy bet or a brute-force bet? Those have different shapes and they need different structures. A brute-force business, especially, wants to be capital-light for as long as possible, because every future financier is going to re-underwrite the same unresolved question, and you do not want to have poured money into a fight that a court can end in an afternoon.

This is Kerr Neilsen’s principle again, the one I keep coming back to. Kerr was BT’s star fund manager when I was starting out and later Australia’s first fund-manager billionaire. He once told me to stop being idealistic about a problem I could not solve and just work out what was going to happen. I wrote about it in the context of Australia’s mineral processing decline, in Don’t be idealistic. This advice applies just as hard to regulatory risk. Do not be idealistic about what ought to happen to the grey-area business. Have a view on what will.

Which brings me to the uncomfortable corollary. In a grey-area industry, the good guy often loses. The founder who waits for every approval, who refuses the aggressive tactic on principle, who does it all by the book, is frequently outcompeted by the one who does not wait. I will never tell a founder to do something illegal. But I will tell a founder they are in an industry where somebody is going to operate in the grey, and it may as well be understood that this is the terrain. The old saying, “Good guys finish last” more often rings true. Sometimes, when I am looking at a clean operator in a dirty industry, I can already see they will be beaten by someone who moves before the rules are settled. That is the same falsification test I described in If I can’t articulate why it will fail, pointed at the ethics of an industry rather than the mechanics of a company.

What this looks like on the deal table

There’s a recent example closer to home, here in Australia, and it is something of a regulatory fork.

Eucalyptus built a large online weight-loss business through its Juniper and Pilot brands. When demand for GLP-1 drugs (the Ozempic and Wegovy class) ran ahead of supply, it prescribed off-label and then went a step further switching to compounded semaglutide, a made-to-order version that was legal to produce while the branded drug was in shortage. That was the unregulated path: cheaper, faster, controlled supply. The same thing was happening in the United States, where a whole industry grew inside the shortage exception at roughly US$150 to $300 a month against branded pricing north of a thousand.

Then the door closed. Australia banned pharmacy compounding of GLP-1 drugs from October 2024, and the US FDA declared the semaglutide shortage resolved in early 2025, shutting the compounding pathway for good. A business built entirely on that exception was now one regulatory determination from nothing. Eucalyptus navigated it, moving patients onto the branded drugs and carrying on. In February 2026 it agreed to sell to the US-listed Hims & Hers for A$1.6 billion, around US$1.15 billion, in a deal expected to close mid-year. It is one of the largest digital-health exits this country has produced. As it happens, the buyer was expanding internationally, partly to diversify away from its own regulatory scrutiny in the US. They had singled out Eucalyptus’s local regulatory expertise as part of what it was buying. Two companies, both reading the regulatory weather, meeting in a deal. That is a regulatory fork done well.

The other fork is ethical before it is legal, and it is the one I find myself raising with founders more often, because it is harder to see coming and it does not stay personal for long. Fantasy sports and sports-data businesses sit next to gambling. The founders tell me, sincerely, that they are not a gambling company and do not want to be lead generation for one. I believe them. It does not matter. The gambling money does not need the founder’s permission to enter: it will copy the product, fund a competitor who has no such reservations, or engage the business whether it likes it or not. The founder’s reluctance is not a moat.

And here the ethical question turns into a political one, which is the whole point. Where a society lands on gambling is not fixed, it moves, and right now it is moving fast, with the rules on what counts as a bet and what counts as a game being rewritten in real time. So the founder who wants to stay on the clean side of that line is making a bet on where the line will be, which is the same bet as the GLP-1 founder reading the regulator and the same bet as xAI reading the administration. A personal moral position is not enough, because the position that feels safe today can be on the wrong side of the politics tomorrow, in either direction. You must have a view on where society is going to draw the line, and when. Which is why, when the ethical gap is real, I want the business capital-light enough to survive the line moving, and sometimes I pass regardless, because I can already articulate the way this one fails. None of it is new, of course. In the 1990s the first businesses to make real money on the internet were gambling and adult content. The medium matured enormously. Those two did not go anywhere. Some grey areas do not resolve. They just mature.

The read I have, and the fork I can’t call

I have said before that I am completely convinced AI changes everything, and I mean more than drafting emails and summarising meetings. That is the first-order stuff, the visible layer. The parts that matter are the second, third, and fourth-order effects, creating the cascade nobody prices at the start (which is the argument I made in After the Machines). Those effects run through energy, through supply-chain security, through politics, through sovereignty, and, at the far end, through war. Once you see AI as the first cause of that chain rather than a productivity tool, the turbines in Mississippi stop looking like an environmental story and start looking like a sovereignty one.

About those turbines, I want to be honest about the cost rather than be clever about it. There is a real community downwind of that data centre, and a real argument that they are wearing the pollution so a supercomputer can train faster. I am a techno-optimist; I do not pretend the journey is clean. Pretending it is clean is just idealism wearing a different hat, and idealism does not change what happens next.

The geopolitics wins. Pandora’s box is open, globally, and nothing is going to put AI back in it. No country gets to opt out, because opting out does not stop the thing; it is simply losing the contest to whoever did not opt out. And the forcing functions underneath that contest are the oldest and strongest there are: sovereignty, and the potential for war, expressed through energy and the supply chains that feed it. When AI capability moves inside that security perimeter, which the US Justice Department has essentially stated it already has, the environmental-first path stops being a settled commitment and becomes a peacetime luxury that holds only until security forces the issue. So, the path snaps. It was always a question of ‘when’, waiting to see how acute the pressure gets and how much a country is willing to give up before it yields.

This is the point of divergence between the USA, Europe, and Australia. The United States has, in effect, already snapped, building first and litigating later. Europe is the hardest to read, because Europe has the most to concede: the deepest environmental commitment, the most bitter pills to swallow to change course. I find it very hard to see Europe walking the US path any time soon, which is not the same as saying it never will. Australia is at its own fork, and honestly, I do not have a strong view. We are temperamentally a country that queues for the permit before it pours the concrete, and on the energy-and-data-centre question I do not see us trucking in the turbines yet. But the same country produced Eucalyptus, a business that read a grey area, operated in it, and navigated the door when it closed. The appetite is there at the company scale. Whether it arrives at the national scale, and when the path snaps if it does, I cannot call.

For an angel, that is the whole lesson in miniature. When the direction is knowable but the timing is not, what you are really underwriting is the when. You are pricing a clock you cannot read, which is why you structure the exposure to survive being early: capital-light, able to wait for the snap rather than betting the business on its arrival by a particular quarter. Reading the politics is reading that clock. It is not a distraction from the investing. It is the investing.

Richard Moore — MooCoo Ventures

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