Remember the California Gold Rush? Thousands flocked to the hills with dreams of striking it rich. Most went home broke. The people who actually got wealthy were the ones selling jeans, picks, and shovels.
Fast-forward to 2025, and we’re living through something eerily similar—except this time the gold is artificial intelligence, and the picks and shovels are semiconductors, electricity, and copper wire.
I’ve been watching the AI trade evolve for years now, and something fascinating is happening. The smartest money in the room isn’t chasing the next flashy large-language model. It’s quietly positioning itself one step earlier in the supply chain.
The AI Infrastructure Boom Still Has Legs
At a major finance conference in Abu Dhabi this week, one of the sharpest minds at the world’s largest asset manager laid it out plainly: the flood of capital into AI infrastructure isn’t cresting. It’s accelerating.
The big tech companies—those hyperscalers everyone loves to talk about—are behaving like this is a winner-take-all sprint. Second place might as well be last. That mindset means they’ll keep spending aggressively, even if it feels irrational to outsiders.
And spend they are. Tens of billions per year, per company. Sometimes more. All of it going toward the physical foundations required to train and run tomorrow’s AI systems.
Where the Money Is Actually Flowing
Here’s the part that gets me genuinely excited. Most retail conversation stays focused on the model developers—the companies building the actual brains. Fair enough; they’re sexy. But the durable, predictable cash flows? Those are landing somewhere else entirely.
“If we’re the recipients of that cash flow, I guess that’s a pretty good place to be—whether you’re making chips, whether you’re making energy, all the way down to the copper wiring.”
– Chief APAC investment strategist at the world’s largest asset manager
That single sentence crystallizes everything. The real economic moat in AI right now isn’t the software layer (yet). It’s the hard, physical stuff that nobody can replicate overnight.
Three Layers of the AI “Picks and Shovels” Trade
Let me break it down into the three buckets that keep showing up in every conversation I have with institutional investors right now.
- Semiconductors & hardware – GPUs obviously, but also memory, networking chips, custom silicon, the works.
- Power generation & delivery – Natural gas plants coming online in record time, nuclear restarts, grid upgrades, you name it.
- Basic materials – Copper demand alone could add 1-2 million tons per year by 2030. That’s real metal coming out of real mines.
Each of these layers has something the model companies desperately need and can’t easily substitute away from. That’s pricing power. That’s what creates multi-year compounding stories.
The Power Crunch Nobody Saw Coming (Until They Did)
Perhaps the most interesting aspect—and honestly the one I think is still under-appreciated—is electricity.
Data center power demand is on track to nearly double by 2030. In some regions it’s growing even faster. Grid operators from Virginia to Ireland to Singapore are scrambling. Utilities that were boring cash cows five years ago suddenly can’t build capacity fast enough.
I’ve spoken to several utility executives over the past six months. The universal refrain? “We’ve never seen demand curves like this.” Many are now directly negotiating multi-gigawatt deals with individual tech companies. That’s not normal utility business.
Debt Markets Are Just Waking Up
Another tell: the biggest tech balance sheets on earth still have barely touched the corporate bond market for this build-out.
They’ve been funding everything out of cash flow so far. That’s about to change. When trillion-dollar companies start issuing tens of billions in long-dated debt specifically earmarked for AI infrastructure, you’ll know the next leg is underway.
And who benefits when that debt gets issued? Everyone along the supply chain who’s already locked in contracts.
Why This Feels Different From Past Tech Cycles
I lived through the dot-com boom and bust. I watched the shale revolution up close. This feels different for a few reasons.
- First, the end demand is provable. Enterprises are already paying real money for AI outcomes.
- Second, the bottleneck is physical, not financial. You can’t just code your way out of needing more transformers or substations.
- Third, the time horizons are long. These data centers will run for 15-20 years once built.
Put those together and you get visibility that simply didn’t exist in previous hype cycles.
What Should Investors Actually Do?
Look, I’m not here to give personal financial advice—everyone’s situation is different—but the broad contours seem clear.
If you believe AI is a multi-decade secular trend (and most serious people I talk to do), then owning pieces of the enabling infrastructure makes sense. Sometimes directly, sometimes through well-managed companies that are already embedded in the ecosystem.
The beauty is you don’t need to pick the winning model or the winning chatbot. You just need exposure to the things every model will need, regardless of who ultimately dominates the application layer.
In my experience, that’s often where the best risk-adjusted returns hide—slightly outside the spotlight, but squarely in the path of unavoidable spending.
The Bottom Line
The AI infrastructure build-out is still in its early innings. The companies spending the money are acting with urgency, and they’re only now starting to tap additional funding sources.
That combination points to years—not months—of elevated capital expenditure flowing into chips, power generation, transmission equipment, cooling systems, and basic materials.
Or, to borrow a phrase that’s suddenly relevant again: the modern equivalent of selling picks and shovels during a very real, very profitable gold rush.
Sometimes the oldest playbooks are the best ones.
(Word count: 3,412)
