The Simplest AI Bet Isn't a Chip Stock. It's the Gas Nobody's Pricing.

What if the cheapest way to own artificial intelligence isn't a chipmaker at all? Fred Wilson, co-founder of Union Square Ventures, thinks it's energy. His argument is blunt and hard to argue with on the surface: natural gas fires most of the new data centers going up across Texas, Virginia, and Georgia, and yet the price of gas has barely budged. The AI trade, in other words, is running on a fuel that the market refuses to reprice.

A Number That Won't Move

Start with the chart. Henry Hub front-month gas has spent most of this year bouncing between roughly $3 and $4 per million British thermal units, with a brief dip below $2 back in early 2024. That's the same range it's traded in for years. No breakout. No squeeze. No panic.

Now the demand side. Data centers accounted for about 4.4% of US electricity consumption in 2023, according to Lawrence Berkeley National Laboratory. By 2028 that figure lands somewhere between 6.7% and 12%, depending on how fast the buildout goes. Gas-fired plants supply roughly 43% of American power generation, and they're the fastest thing to permit and pour concrete for. So the demand shock is real, documented, and already funded.

Chip stocks, meanwhile, have taken a knock this month. That's the contrast Wilson is pointing at. The picks-and-shovels trade got crowded and expensive. The fuel underneath it didn't.

Why Gas Stayed Boring

Here's the part the bulls gloss over. US dry gas production is sitting near record territory, north of 100 billion cubic feet a day, and shale operators have gotten uncomfortably good at responding to price signals. Associated gas out of the Permian keeps showing up whether anyone asks for it or not.

So demand grew, supply grew faster, and the price went sideways. That's not a market ignoring a thesis. That's a market pricing in elasticity. History suggests otherwise when supply is genuinely constrained, and gas supply hasn't been constrained in years.

Which is why the commodity itself is probably the worst way to express this view. Gas is a volatile, weather-driven, storage-sensitive contract that can drop 20% on a mild February. Proponents of the Wilson thesis are right about the direction and wrong about the vehicle. Buying futures gets you whipsawed. Buying the constraint points gets you paid.

The Bottleneck Is the Trade

Color me skeptical, but I'm not entirely convinced the energy takeaway is really about the fuel at all. It's about everything between the wellhead and the server rack, where supply can't expand on a two-year timeline.

Gas turbines are sold out into the back half of the decade. GE Vernova's order book has been the tell for two years running. Transformers are on multi-year lead times. Interconnection queues in PJM and ERCOT stretch out past 2027. Pipeline permitting still moves at the speed of paperwork, and no amount of demand changes that.

Those are the assets with pricing power. A driller selling a commodity into a record-supply market has none.

The question worth asking: if AI demand is as large as everyone claims, why hasn't it moved the one input every projection says it needs most? Either the demand forecasts are too generous, or the supply response is better than the skeptics admit. Both answers argue against a simple long-gas position.

What to Watch

The near-term calendar is honest about this. The EIA publishes its weekly storage report every Thursday, and the injection and withdrawal numbers against the five-year average will tell you more about gas direction than any AI press release. Watch the $4 per MMBtu line on Henry Hub. A sustained break above it in a mild winter would be the first real sign that AI load is showing up in the physical market.

Then there's LNG. Golden Pass and the next wave of Gulf Coast export capacity add a second demand pull that has nothing to do with data centers, which muddies the read on any AI-specific story. And Nvidia's next earnings print, expected in late November, will reset the whole adjacent trade either way.

Wilson's framing is smart because it forces people to look past the obvious. Just don't confuse a good observation with a good ticker. Time will tell, though I'd rather own the turbine backlog than the gas contract.