Some of the most-cited figures in AI infrastructure coverage are also the least reliable. Projections of $3 to $4 trillion in cumulative AI infrastructure spend over the coming years come from sources with opaque or inaccessible methodology. They’re not in this brief.
Here’s what the numbers that can be defended actually show.
Bridgewater Associates, in analysis reported by Reuters, estimated that the four largest hyperscalers, Alphabet, Amazon, Meta, and Microsoft, were on track to spend approximately $635 to $665 billion combined in their 2026 fiscal years. That figure’s primary source is a broken URL in available research; the recommended framing is: Bridgewater Associates estimated the Big Four were on track for that range, per Reuters reporting. TechCrunch summarized the Bridgewater estimate in March 2026.
Goldman Sachs, in its own published analysis, projected that AI hardware spending by major cloud providers could surpass $500 billion in 2026. Goldman’s analysis represents the firm’s own forward estimate, not a third-party audit. Directionally, it corroborates the scale suggested by the Bridgewater figure.
Both estimates, taken together, describe a capital commitment with few historical comparisons in the commercial technology sector.
Now the constraint.
TechCrunch reported in March 2026 that power availability has become “one of the biggest bottlenecks in rolling out new AI data centers.” That framing reflects a structural reality: the build-out’s pace is no longer primarily limited by capital availability or hardware lead times. It’s limited by how fast grid infrastructure, power purchase agreements, and permitting processes can keep up.
A BlackRock investor survey, reported by Reuters, found more than half of respondents preferred energy infrastructure companies over hyperscalers for AI investment exposure in 2026. That Reuters report, as referenced in TechCrunch’s March 2026 energy-infrastructure coverage, reflects a meaningful shift in where sophisticated capital is positioning. It’s not that investors are abandoning hyperscalers. It’s that the energy constraint has made the companies solving the power problem look like a better risk-adjusted bet in the near term.
This creates a specific scenario worth watching: the hyperscalers have the capital to build. They’ve committed it. The build-out’s timeline is now governed by energy availability, not balance sheet capacity. That changes which companies benefit from the AI infrastructure wave.
What to watch: power purchase agreement announcements from Alphabet, Amazon, Meta, and Microsoft; regulatory approvals for new data center campuses in energy-constrained markets; and whether energy infrastructure companies begin to appear in the same funding and M&A cycles as the AI companies they’re being asked to power.
The capital story is settled. Six hundred and fifty billion dollars is committed. The next story is whether the grid can absorb it.