Oracle's Project Jupiter and the Real Bottleneck
Oracle (ORCL) Project Jupiter shows the problem in AI is not demand. The problem sits in the infrastructure itself. A permit for a natural gas pipeline stalled. That gas was needed to power some of the fuel cells, and without the permit the project slipped.
Wall Street read the delay wrong. Oracle (ORCL) structured the deal so it could push back payments tied to any hold-ups in launching data centers. This was a financial choice, not a sign of weak demand. It was built around a timed, planned launch. So the worry is not that capacity is growing too large. It is a straight infrastructure bottleneck: the pipes, power, and buildings needed to run these data centers will get built, but that takes time and resources.
Funding Is Not the Constraint
Private credit and equity sit at the center of financing this build-out, and that corner got hit hard. Blue Owl (OWL) sits inside this story on the private credit side, and its private equity fund is struggling with rising rates, which shifts the math somewhat. Many private companies now borrow from private lenders instead of going public.
Still, financing will not be the wall. Big tech companies hold huge piles of cash and carry clean balance sheets, much cleaner than their large-cap rivals. They will have no trouble paying for this. The real test is how fast they can put that money to work and clear the logistics of rewiring the whole network and building the ecosystem around it. Those steps matter more than raising the cash. Money is available, and money always follows good, profitable ideas.
Small and Mid-Cap Picks and Shovels
The smarter play is owning the picks and shovels rather than guessing which AI firm wins. Every data center needs energy engineering firms that know how to build these things and grow them from an early stage into a full structure.
Emcor (EME) - market cap of $33 billion, trades near 22 times forward earnings, with a strong cash flow yield near 30%. These are the players building the data centers.
Belden (BDC) - a smaller company near $4 billion. It sits at the intersection of the routers and switches that link data centers and let cloud infrastructure roll out. It trades at 12 times, has a book value ratio above one, and a profit margin that could grow to 15 to 20%.
Powering It All With Natural Gas
By 2027, commercial use of natural gas will pass residential use for the first time in a long stretch. The cause is all the new data centers and AI running on them. Natural gas is likely the cleanest and most efficient way to feed that load.
National Fuel Gas (NFG) could gain here. It is filed under utilities but is really a natural gas exploration and production company. It yields 3%, trades at 10 times price, and its earnings are likely growing around that level. Gas is not only feeding data centers. The US now exports more of it than ever, shipping liquefied natural gas to Japan and Europe, where gas prices run far above US levels.
The Energy Bottleneck: Onshoring vs. AI
Onshoring and AI now fight over the same infrastructure. After COVID, there was a push to bring manufacturing back to the US for national security reasons - semiconductor chips from Taiwan, medical devices from China. Heavy money is going into building that manufacturing at home, and those plants will burn a lot of energy. AI growth plus reshored manufacturing together will consume enormous power, and gas is what runs it.
Battery storage is also needed. Some of the alternative grid sources run unevenly, so storage keeps the supply stable and reliable over time.
Early Innings
We are in the early stages of AI. There is far more room ahead, especially once you count consumer adoption and the compute needed to handle all this data and processing. If the macro picture cooperates - rates now at 5.25%, with some cap on further increases - the runway is long.
The top seven S&P 500 companies are set to spend about $1 trillion on capital expenditures in 2027, close to what the entire S&P 500 spends. Seven firms are laying out that much with no leverage involved, and they are generating strong cash. That should keep running for a good while. Naming the AI race winner is anyone's guess, which is why owning the picks and shovels is the better bet. This build-out supports GDP and the broader economy, and it is still just getting started.
