The Appraisal, April 2026 — The State of Data Centers in 10 Charts
This is a special edition of the newsletter on the data center development boom. Data centers are, at their core, a real estate story. They require land, permitting, construction, and long-term leases — and the markets where they concentrate tend to see ripple effects across industrial, office, and energy infrastructure. Let’s dig in!
AI is driving one of the biggest infrastructure build-outs in modern history. Meta, Alphabet, and Amazon are now collectively funneling more than a third of their revenue into capital expenditures, driving data center development to more than 1.2% of U.S. GDP. Whether this is rational catch-up to surging demand or the early signs of an infrastructure bubble — think 1990s telecom or 1870s railroads — is a hotly debated topic. Here are 10 charts that help make sense of it.
1. Historic capital deployment into AI infrastructure
Big Tech is pouring capital into data center infrastructure at a pace with no modern precedent. As a percentage of total economic output, this spending exceeds the peak of the 1990s telecom boom. And for the first time, data center construction spending has nearly eclipsed general office construction — $42.1B vs. $40.5B annually.
The capacity pipeline tells a similar story. In 2010, the U.S. had less than 1 GW of total data center capacity. It took the better part of a decade to reach 4 GW. Then the curve bent — 5.5 GW in 2020, 9.2 GW in 2022, 11 GW in 2023. By 2025, there are 34 GW under construction and 88 GW in the planned pipeline.
Real private fixed investment in data centers has nearly tripled since 2020, while every other category of non-residential construction has barely moved — indexed to Q1 2020, data centers are at 295; everything else is at 98. The two lines, which tracked each other for years, have completely decoupled.
2. Demand for compute is accelerating
The obvious question is whether anyone actually needs all of this. So far, the answer keeps coming back yes.
AI is monetizing faster than any prior technology wave. Cloud revenues across AWS, Azure, and Google Cloud accelerated in 2025. Aggregate revenue is approaching $100 billion per quarter and is expected to continue growing at roughly 26–30% through 2026 — with no meaningful deceleration.
The major AI labs are reporting remarkable revenue growth — Anthropic announced that it had surpassed $30 billion in annualized revenue this month, up from $19 billion the month before and from $1 billion just fifteen months earlier. OpenAI is on pace to surpass $25 billion in revenue this year.
Usage data tells the same story. Token consumption on OpenRouter hit 7.6 trillion per week in Q4 2025. GitHub hit 1 billion commits in all of 2025; it’s now running at 275 million per week, on pace for 14 billion this year if growth holds.
At the peak of the dot-com bust, most of the fiber that had been laid sat completely unused — so-called 'dark fiber' that wouldn't see real demand for another decade. Today's GPU utilization tells the opposite story. AI data centers are running at 80% capacity, and operators are struggling to bring supply online fast enough to keep up. The hyperscalers are consistently reporting that AI workloads are outpacing available capacity.
3. Power is the bottleneck
Despite historic capital deployment and demand that shows no signs of plateauing, a hard constraint is emerging — and it has nothing to do with chips or capital. Getting power to a data center has become the defining bottleneck of the AI infrastructure build-out.
New data center project announcements dropped nearly 50% last quarter — not because demand dried up, but because securing a power connection has become nearly impossible. Developers aren’t starting new projects; they’re focused on finishing the ones already in the queue. The bottleneck has moved from “can we fund it” to “can we plug it in.”
A survey of 149 senior data center professionals lays it out plainly. 92% cite utility or transmission access as their primary constraint. Permitting is second at 86%. Fiber availability third at 85%. The list reads less like a construction industry survey and more like a report on failing infrastructure.
To work around it, 62% are building on-site power generation — fuel cells, turbines, solar — essentially becoming their own utilities. Another 40% are relocating projects entirely to wherever the grid has room.
The downstream implications are significant. Data centers consumed 3.7% of total U.S. power demand in 2023. McKinsey forecasts that figure hitting 11.7% by 2030. To put that in context: the entire U.S. residential lighting load is roughly 5%. In seven years, AI infrastructure alone could draw more than twice that.
Thanks for reading! If you have any colleagues that may find The Appraisal valuable, please share this newsletter.
Note: Charts and graphs provided within are for informational purposes solely and should not be relied upon when making any investment decision. The content speaks only as of the date indicated. Any projections, estimates, forecasts, targets, prospects, and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others.