Jensen Huang has said, on the record, that Nvidia is sitting on at least $1 trillion in confirmed purchase orders through 2027. Not analyst guesses. Not a rosy forecast slide. Confirmed backlog. That single number is doing more to move markets right now than almost anything else in tech, and it's worth understanding both halves of the story: why the number is real, and why the thing standing in its way isn't a competitor, a chip shortage, or a recession. It's the electrical grid.
The Demand Side Is Not Hype
Start with what's actually happened, not what might happen. Nvidia's data center segment pulled in $75.2 billion in a single quarter (the quarter ended April 26, 2026), up 92% year over year. Zoom out further and the growth is almost hard to believe: data center revenue went from $2.98 billion in fiscal 2020 to $193.7 billion in fiscal 2026, a 65x increase in six years. That's not a story being told by Nvidia's marketing team. That's a company's actual reported financials.
Behind that revenue is a small group of buyers spending at a scale that's genuinely new. Microsoft, Amazon, Google, and Meta, the four biggest hyperscalers, are each projected to spend well over $30 billion on data center infrastructure in 2026 alone. That concentration is exactly why the backlog number carries weight: it isn't spread across thousands of small customers who might quietly cancel, it's a handful of the best-capitalized companies on Earth racing each other to build AI infrastructure before their competitors do.
The Real Risk: Nvidia Can Build Chips Faster Than the Grid Can Deliver Power
Here's the part of the story most hype-driven coverage skips entirely. In Santa Clara, California, the city where Nvidia is headquartered, there are data centers sitting fully built and empty right now, waiting for the local utility to be physically able to supply them with power. That's not a funding problem or a chip shortage. It's a hard infrastructure ceiling.
The scale of the problem is bigger than one city. Data centers used about 4.4% of total U.S. electricity in 2023, and that share is forecast to climb to nearly 12% by 2030. Utilities can't add power plants, substations, and transmission lines fast enough to keep pace, and the public is starting to notice. A Gallup survey from March 2026 found roughly seven in ten Americans oppose new data center construction in their own area, largely over fears of higher electricity bills and strain on local water supplies used for cooling. That opposition already has teeth: at least 75 data center projects worth a combined $130 billion were blocked or canceled in just the first three months of 2026, and multiple states have paused new approvals pending grid-capacity audits.
None of this changes whether Nvidia can design and manufacture the chips hyperscalers want. It changes how fast those chips can actually be plugged in and put to work, which is a completely different bottleneck than the one most coverage of the "AI trade" focuses on.
Why This Matters More Than a Normal Supply-Demand Story
In a typical hardware cycle, a backlog this size would mostly be a manufacturing and logistics question: can the company make enough units, fast enough, without a competitor undercutting it. Nvidia's story adds a layer most stocks never have to deal with: local government decisions, thousands of miles from its headquarters, in cities that have never previously mattered to its business, can now directly delay revenue that's already been ordered and paid a deposit on. A data center that can't get power isn't a canceled order, but it is deferred revenue, and deferred revenue at this scale is a real variable, not a footnote.
What to Actually Do With This
- Separate the two questions. "Is AI infrastructure demand real?" and "Will that demand convert into revenue on the timeline the backlog implies?" are different questions with different answers right now. The first is genuinely, verifiably yes. The second has a real, physical bottleneck attached to it that has nothing to do with Nvidia's own execution.
- Watch grid and permitting news, not just earnings calls. A state pausing data center approvals or a utility publicly stating it can't add capacity is directly relevant information for this story, even though it never appears in a traditional stock screener.
- Don't treat a single company's fortunes as a bet on "AI" broadly. If you want exposure to the theme without picking a single winner, a broad index fund already owns pieces of the hyperscalers doing the spending, the chipmakers selling to them, and everything in between, without requiring you to correctly guess how the power-grid bottleneck resolves.
The demand side of this story is as close to verified fact as investing narratives get. The power-grid side is the part almost no one is pricing in correctly yet, in either direction. That gap between the two is exactly why this is a story worth following past the headline number.
—
Spicy Investing