Who Keeps the Gold?

The argument right now is whether we're building too much AI power or too little. In a recent report, Morgan Stanley says there's a shortfall. McKinsey, on the other hand, says we're overbuilding. Both could be right. Either way, the headlines are full of grid queues and stalled projects.

I follow all of it and I keep coming back to one thing: both sides start from the same assumption. It's the one I'd question.

Is owning the power the ultimate prize?

The instinct behind that is an old one. In a gold rush, don't dig, sell the shovels. Supply the people doing the gambling and let them carry the risk. Applied to AI, the same instinct says that the safe money is in the tools. The power, the data centers, and the chips. That is where money is going right now, and on the surface, it looks sensible.

The trouble is, most shovel sellers don't keep the gold. NVIDIA does, but only because its moat is genuinely hard to copy. Go back a generation, to the companies that laid the broadband cables. They sold the essential tool of their moment, then spent the next twenty years running a commodity on thin margins while the real money moved to whoever built on top of the pipe. Same instinct, opposite result.

So the question isn't tools versus users. It's whether the scarcity lasts. If a few years of building can erase the advantage, it's temporary. If it's rooted in something hard to copy, like a location, a long-term contract, or a real regulatory or grid barrier, then it's durable. The first gets competed away. The second holds.

Measured that way, a lot of today's AI power looks more like the broadband cable than the moat. Generic megawatts and generic data centers are priced as if they'll stay scarce, while the whole industry races to build more of exactly them. The demand is real. The IEA reports that the electricity data centers consume is rising fast, and the share going to AI is rising fastest. Gartner expects data centers to use more than a quarter more electricity again this year. And according to the World Economic Forum, getting all that new capacity connected to the grid is the real bottleneck. These are all true but the scarcity that everyone is racing to build more of will likely not stay scarce for long.

The user side is where I'm less sure, and I'd rather say so than pretend otherwise. History says value migrates to whoever builds on the commoditized layer. But AI has a twist the broadband era didn't. If the tools end up in more or less everyone's hands, then using them isn't obviously an edge. It might just be the cost of staying in the game. The company that deploys AI well could look up the next day and find that its competitors have done the same. And just like that, the customer, not the provider, kept the gain. Moat or the price of admission? I don't think the answer is the same in every industry, and I don't think anyone knows it yet.

There's an old idea in strategy I keep coming back to. When one layer of a business turns into a commodity, the profit doesn't completely vanish. It moves onto the next layer, wherever scarcity still lives. If that holds, the winners in AI on either side of the picks-and-shovels line will be whoever owns something AI can't hand out. AI makes what you already have more valuable. It doesn't create it for you.

So who wins? Here's where I land. Not the people buying the tools and hoping they stay scarce. The picks and shovels get cheaper every year, and before long everyone can swing them. The ones who keep the gold will be sitting on something AI can't manufacture. Own that, and AI makes you stronger. Own the pipe, and you're the broadband company again, waiting to be turned into a utility.

Plainly: a lot of the money going into AI power right now is paying for scarcity that won't last. I'd back the thing the scarcity can't be built around.

So what does that look like in practice? A few places I'd keep my eyes on.

Proprietary data. Not the open web everyone can scrape, but the regulated health, financial, and industrial datasets only a handful of players hold. AI is only as good as what you feed it, and that data isn't for sale.

Locked-in customers. The incumbents whose users can't easily walk. AI makes a relationship you already own stickier. It rarely hands you a new one.

Location. Power will be won and lost by geography. Cheap electricity, a grid that can actually connect you, and a regulator who wants you there beat a bigger check somewhere else.

The risk I'll own: I could be wrong on timing. If this scarcity holds up longer than I expect, the people who own the power today will do very well for a good while first. I'm more convinced about the direction than the timing.

You might see it differently, and I'd want to hear why. But if you're writing a check into any part of this, the question isn't how much power you're buying. It's whether you're paying for an edge that lasts, or one the next few years will erase.

The views expressed here are the author’s own and do not constitute investment advice or a recommendation to buy or sell any security. See the full disclaimer below.