Finance Studio Advisors · The Ledger Letter
Wall Street Spent Two Years Chasing the Chip. Now It Has to Power It.
The four largest hyperscalers just committed up to $725 billion in capital spending for 2026, per the Financial Times. That is 77% more than last year’s record. Nearly all of it flows into AI data centers. But half the data center capacity planned for this year in the United States is stalled, per Bloomberg. Stuck behind grid bottlenecks, transformer shortages, and transmission queues stretching five years deep. The market priced AI like software. The bill is arriving like infrastructure.
The Breakdown
01 The Money Is Moving Underground
Of the roughly 12 gigawatts of U.S. data center capacity announced for 2026, only about 5 GW is under active construction, per Bloomberg. The rest sits frozen. Not by chip shortages. By electrical infrastructure: transformers, switchgear, grid connections. Lead times for high-voltage transformers have stretched from 18 months to as long as four years.
02 The Grid Was Never Built for This
Data centers accounted for roughly half of all U.S. electricity demand growth last year, per the IEA. In Virginia alone, they consumed nearly 40% of the state’s total electricity in 2024, per Bloomberg. Last July, 60 data centers in Northern Virginia disconnected from the grid simultaneously after a single equipment failure. The load that vanished in seconds: 1,500 megawatts. That is more electricity than all the households in some mid-sized U.S. states use at peak.
03 The Race Is Leaving the Ground
SpaceX filed with the FCC in January for an orbital data center constellation of up to one million satellites, per Data Center Dynamics. The filing projects 100 gigawatts of AI compute capacity annually. For context, that equals roughly 20% of current U.S. electricity consumption. Amazon’s Project Kuiper faces an FCC deadline to deploy 1,618 satellites by July 2026. The constraint on AI scaling has become physical. The companies solving it may not be the ones building the models.
Read the capital flow, not the headline.
For eighteen months, the AI trade was a chip trade. NVIDIA rose. Broadcom rose. The Philadelphia Semiconductor Index ran for 18 straight sessions in April. The logic was clean: whoever makes the GPU wins the AI race. That logic was correct. It was also first-order.
The second-order question is harder. Who powers the GPU? Who cools it? Who connects the data center to the grid, and who builds the grid to handle the load?
The U.S. interconnection queue now holds roughly 2,600 gigawatts of projects, per Lawrence Berkeley National Laboratory. That backlog exceeds the entire installed generating capacity of the American power grid. The median wait time for a new project to reach commercial operation is approaching five years. Morgan Stanley forecasts a 49-gigawatt shortfall in U.S. data center power by 2028.
The semiconductor trade assumed an infinite power socket. The socket has a line.
By the Numbers
AI capex vs. the grid that carries it.
| Metric | Figure |
| 2026 Big Four hyperscaler capex | $725B |
| Year-over-year increase | +77% |
| U.S. data center capacity announced for 2026 | ~12 GW |
| Actually under construction | ~5 GW |
| U.S. grid interconnection backlog | 2,600 GW |
| Morgan Stanley projected U.S. power shortfall by 2028 | 49 GW |
Sources: Financial Times, Bloomberg, IEA, Lawrence Berkeley National Laboratory, Morgan Stanley. Figures reflect latest available data as of May 2026.
Partner Perspective
|
Editor’s Note: Former tech executive and angel investor Jeff Brown picked Bitcoin before it jumped as high as 52,400%, Tesla before it jumped as high as 2,150%, and Nvidia before it jumped as high as 32,000%. Today, he’ll show you how to claim a stake in Elon Musk’s upcoming IPO — BEFORE the company goes public. Click here to see the details or read more below. Dear Reader, Over the past few weeks, I’ve been urging my readers to claim their stake in what I believe to be the biggest IPO of the decade. And I’m glad I did. Because over the last 21 days, three critical events happened in rapid succession: ✓ March 17th: SpaceX crossed 10,000 active satellites in orbit. The estimated threshold for offering full service to most of the globe. Two-thirds of every satellite circling Earth now belongs to ONE company. ✓ April 1st: Elon filed the confidential IPO paperwork with the SEC. The public filing could drop any day now. And when it does, the stampede begins. ✓ April 6th: Another rocket launched carrying 25 more satellites. Proving SpaceX isn't slowing down. They're accelerating. Building the network that will become the world's first global internet carrier. SpaceX just hit every technical milestone it needed to justify going public. Everything I predicted is happening... right on schedule. And there's still a small window to get in BEFORE the public can buy shares. But that window is closing fast. The moment the public filing drops, millions of investors will learn about this opportunity for the first time. You won't be early anymore. You'll be competing with the crowd. And your shot at early gains will be gone forever. See how to claim your stake in SpaceX before it’s too late. We have so much to look forward to, Jeff Brown |
The Full Picture
AI Stopped Being a Software Story. The Capital Cycle Just Noticed.
The crowded trade
The semiconductor position was correct. It was also first-order. NVIDIA’s run made the logic feel obvious. Build the chip, win the race. But the second-order question was always lurking beneath it: who powers the chip once it ships?
The U.S. interconnection queue now holds roughly 2,600 gigawatts of projects, per Lawrence Berkeley National Laboratory. That backlog exceeds the entire installed generating capacity of the American power grid. Google guided up to $190 billion in 2026 capex. Microsoft matched it. Amazon held at $200 billion. Their cloud backlogs are swelling. Google Cloud alone reported $460 billion in backlog, roughly double late 2025. The demand signal is real. The delivery mechanism is cracking under its own weight.
The cross-asset read
Here is the tension. Sightline Climate’s February 2026 analysis found that nearly 50% of global data center projects scheduled for this year face delays tied to power supply limits. Alphabet’s free cash flow may fall nearly 90% this year, per Pivotal Research, as capital pours into physical build-out. The pipeline of conditional agreements between data center operators and small modular nuclear reactor projects has grown from 25 gigawatts in late 2024 to 45 gigawatts today, per the IEA.
Our view: the capital rotation already underway is not away from AI. It is from AI-as-software toward AI-as-infrastructure. Grid operators, transmission builders, energy storage, satellite connectivity, modular nuclear. These are the categories where capital is beginning to migrate. The investors still holding a pure semiconductor position are watching the right race from the wrong seat.
What this means for your portfolio
Think about the framing differently. When the internet scaled in the late 1990s, the first winners were the software names. The lasting winners were the companies that laid the cable, built the towers, and delivered the bandwidth. AI may be following the same arc at twenty times the speed and fifty times the energy cost.
SpaceX filing for a million-satellite orbital data center constellation is not a curiosity. It is a data point about where the constraint sits. When a launch company starts filing for compute infrastructure in orbit, the terrestrial grid has sent its answer.
Honestly? If your AI exposure is still concentrated in chipmakers and model builders, the position earned its returns. The question is whether the next leg of returns comes from the same layer of the stack. Worth watching: infrastructure names, power generation, transmission capacity, and orbital connectivity. The rotation is quiet. It is already moving.
The next AI winners may not be the companies building the intelligence. They may be the companies carrying it.
Finance Studio Advisors
Precision in every paragraph. Financial communication built for clarity, credibility, and action.
