For nearly two years, Wall Street has been telling itself a beautiful story about artificial intelligence. Light capital requirements. Infinite margins. Winner-take-all economics. All the good stuff.
Jeff Currie's recent analysis for Carlyle is here to ruin that party. His message is blunt: AI isn't just software. It's infrastructure. And infrastructure always drags commodity economics along with it.
The core insight is simple but deeply uncomfortable for investors who've been riding the AI wave. GPU hours are now priced, benchmarked, and competed away in real time. Supply is expanding rapidly. Prices are falling faster than anyone expected. AI compute is starting to behave less like a differentiated technology service and more like, well, a commodity.
If you lived through the shale boom, this should give you serious déjà vu.
During the 2010s, shale producers operated under the assumption that oil prices would stay structurally high. Capital flooded into drilling operations. Companies routinely spent more than their cash flow. Production surged.
The outcome wasn't lasting shareholder prosperity. It was collapsing prices and years of capital destruction. Only after the industry was forced into discipline did energy equities become attractive investments again.
We're watching a similar pattern unfold in AI compute right now.
How to Position Your Portfolio for This Reality
Big Tech is spending money at a scale that would make even shale drillers nervous. We're talking hundreds of billions being poured into data centers, chips, power contracts, and cooling infrastructure. These aren't asset-light investments. They're steel in the ground, long-dated commitments, and fixed cost structures.
Meanwhile, compute pricing is already falling as specialized providers and global competitors flood the market with capacity.
This has serious implications for how you should think about public equities.
For technology stocks, the critical distinction moving forward is between companies that sell AI-enabled products and companies that sell raw compute. The latter increasingly looks like a utility or commodity business: heavy capital requirements, declining unit prices, intense competition. Returns driven by cost curves rather than innovation narratives.
Hyperscalers might still win strategically, but massive AI capital expenditures don't automatically translate into higher margins or durable shareholder returns. History actually suggests the opposite. When an industry builds ahead of demand, customers win early. Capital providers usually don't.
This doesn't mean AI is a bubble. It means the economics are shifting underneath everyone's feet. Software companies that can leverage AI without owning the infrastructure remain compelling. Asset-heavy compute providers should be valued like infrastructure businesses, not high-margin growth stories.
The Energy Infrastructure Angle Everyone Is Missing
Here's where things get interesting. Step outside technology for a moment and look at energy and midstream.
Unlike shale in its early, chaotic days, midstream energy companies today are behaving with actual discipline. Pipelines, gathering systems, and processing plants are being built against long-term contracts. Capital allocation is conservative. Balance sheets are stronger. Returns are visible and real. And crucially, demand is genuine.
AI data centers are power-hungry in ways the market is only beginning to appreciate. Natural gas is the marginal fuel for that power. Midstream companies sit right at the center of this reality. Every new data center cluster needs gas supply, transportation, processing, and storage.
Unlike compute pricing, gas transportation isn't easily commoditized away. Geography matters. Infrastructure matters. Contracts matter.
In other words, AI might compress returns for some technology infrastructure owners while simultaneously expanding opportunities for energy infrastructure owners.
This is the inversion investors consistently miss. The sexy story attracts too much capital. The boring backbone quietly compounds value.
What History Teaches About Capital Cycles
The shale analogy is powerful because it reminds us that capital cycles always resolve the same way. Oversupply creates price pressure. Price pressure forces discipline. Discipline creates opportunity.
In energy, we're already in the disciplined phase. In AI compute, we're still firmly in the build-at-any-cost phase.
For investors, the lesson isn't to abandon AI entirely. It's to be selective. Favor businesses that benefit from AI adoption without bearing commodity risk. Be cautious with companies whose economics increasingly resemble heavy industry priced by the hour.
And don't ignore the second-order winners. Midstream energy isn't just an inflation hedge or a yield play anymore. It's becoming critical infrastructure for the digital economy. Unlike compute, its pricing power is rooted in contracts, regulation, and physical scarcity.
Markets always overestimate the returns of the shiny new thing and underestimate the value of the plumbing that makes it work. That was true in shale. It's proving true again in AI.
As always, the job isn't to chase the story. It's to follow the economics.
