AI's Next Winners Aren't the Companies Getting the Money

In Mount Pleasant, Wisconsin, Microsoft built a $3.2 billion data center campus. The buildings are finished. The cooling systems are installed. The GPU clusters sit in storage, ready to be racked. The facility still will not process a single AI workload until late 2026 at the earliest this facility won't process a single AI workload until late 2026 — not because the software isn't ready, but because the equipment that steps the grid's power down to something a server can handle hasn't arrived.

That image is the whole argument behind a headline now making the rounds: that AI's next winners are the companies receiving the funds. The hyperscalers are handing money out at an unprecedented rate — hyperscaler capex will reach $697 billion in 2026 — and, once you add the borrowing, Morgan Stanley sees roughly $2.9 trillion of data-center investment through 2028, with about a $1.5 trillion gap that credit markets, not equity, are being asked to fill. By that logic, whoever holds the order book with the cash flowing into it becomes the next big winner.

The recipients of that money are real. Quanta Services (PWR), the Houston contractor that builds the high-voltage transmission lines, substations, and electrical "balance of plant" inside data centers, is the cleanest case. Its backlog just hit a record $53.4 billion. Revenue grew 41% in the most recent quarter on $9.56 billion; adjusted earnings of $4.24 a share clobbered a Street expectation near $3.29, and the stock jumped roughly 14% the day it reported. It is up about 47% this year. Quanta is not merely "exposed" to AI — it is on the receiving end of tens of billions of committed dollars.

So the thesis holds at the surface. The problem is that receiving money and capturing profit are not the same business, and Quanta is exactly where the two split.

Being necessary gets you orders. Being scarce decides who keeps the money.

Quanta is a labor company. It fields a craft workforce of roughly 85,000 people and self-performs 80–85% of the work it wins. A contractor bills the customer for that labor, the materials, and a margin on top. That structure is why Quanta's operating margin sits around 5–6%: most of every revenue dollar is a pass-through of crews, steel, and fuel. More volume pushes the top line up impressively — that 41% — without doing much to the profit earned on each dollar of revenue. The free-cash-flow margin actually slipped, to about 5.7%, and Quanta is spending nearly 30% more on capex this year, around $775 million, just to serve the backlog that much faster.

This is the crux. A transformer is scarce because only a handful of companies can make one, and no amount of money shortens the queue. A construction crew is scarce only as a hiring problem — and crews get bid on, project by project. Quanta competes for this work against MasTec and EMCOR and a raft of regional players. Scarcity that can be outbid is not pricing power. Being necessary gets you the order; being scarce is what lets you keep the economics.

Where is the real scarcity in this boom? Not in the labor that hangs the wire, but in the equipment the labor has to wait for. Only five or six companies in the world build large power transformers at scale — GE Vernova, Hitachi Energy, Siemens Energy, ABB, and a couple of Asian suppliers. Lead times for those units have stretched from 24–30 months to three to five years, and supply runs a rough 30% shortfall. Bloomberg has reported that more than half of the U.S. data centers planned for 2026 are now delayed for lack of transformers and switchgear, with only about a third of the 12–16 gigawatts planned this year actually under construction. In Northern Virginia, Dominion Energy is processing more than 60 gigawatts of connection applications against roughly 8 gigawatts of available capacity.

Money runs faster than capacity in every direction here. A hyperscaler can approve capital tomorrow, but it cannot conjure a qualified transformer or a grid-interconnection slot in years that do not exist. That is a genuine bottleneck, and it is why the equipment-makers — GE Vernova, Eaton — carry the pricing power. Eaton trades around 40 times trailing earnings and earns margins several times Quanta's. The name with the biggest number in the backlog is not the name with the pricing power.

The irony that cuts against the biggest recipient

Here is the part that should change how you read that $53.4 billion. The same equipment shortage that makes the AI spend look unstoppable is what gates Quanta's own timeline. The Wisconsin campus does not process workloads because the transformers are not there — work Quanta would otherwise be billing. Finished buildings, GPU clusters in storage, waiting on a box with a three-to-five-year lead time.

And Quanta's own management has said most of its pipeline is not even booked. Of the largest transmission corridors and generation programs the company is pursuing, roughly 95% are not yet in the backlog. Backlog is a promise about contracts already signed; a pipeline is an ambition. At some point a $53 billion order book has to convert — into revenue, then into margin, then into cash — and each step can be delayed by a permit, a transformer, or a lower competing bid.

That is the risk embedded in a staggering price. Quanta trades near 50 times this year's earnings estimates, about double its five-year average. At that multiple, the market is paying years ahead for near-flawless conversion of a backlog the company cannot accelerate, into margins it has not expanded, while cash conversion is being consumed by growth. It is also a fragile multiple: the stock is down about 8% over the past month even after the record quarter, as some capital rotates out of the most expensive names in the AI-capex trade.

None of this says AI spending is a bubble or that Quanta is a bad company. It says the headline has picked the wrong winner. The companies "receiving the funds" are the ones with the biggest order books, and Quanta's $53.4 billion is a genuine, enviable number. But the companies that get to keep the money are the ones selling the step the funds cannot buy quickly: the transformer, the switchgear, the grid connection. There, five suppliers set prices against three-to-five-year waits. In the building layer, dozens of firms bid against each other for crews that can always be hired.

When you next hear "the winners are the companies receiving the funds," translate it into two questions. Does the revenue create pricing power, or is it a pass-through that merely makes the top line enormous? And which step can money not make faster — because that is where the rent lives, and it is rarely the step with the biggest headline backlog.

For Quanta, the operating margin is both the confirmation metric and the escape signal. If margin begins to track the backlog's growth, the premium starts to make sense — a contractor turning record volume into record profitability. The moment that fails, when tens of billions of backlog keep flowing in while the margin will not budge, is the moment this "next winner" is revealed to be a bigger version of an ordinary business — receiving more than it ever captures.