Summary

  • Amazon’s U.S. self-built data centers consume up to roughly 9 gigawatts of power — comparable to North Dakota’s generation capacity — giving it an incumbent lead built on two decades of construction, per data provider Aterio; Microsoft and Google each run about 5 gigawatts of self-built capacity, Meta about 4.
  • Amazon is projected to add the most capacity through 2030, but Google is projected to add capacity at the fastest rate; counting leased third-party capacity, Aterio expects Google to have “significantly closed its gap” with Amazon by 2030.
  • The two leaders are resolving the same constraint with different trade-offs: Amazon’s strategy emphasizes cost and reliability (mostly self-built, longer to stand up, cheaper long-term), while Google’s emphasizes clean energy (heavier on leases, in-house renewables developer, grid-flexible).
  • Both leaders accept fossil-fuel exposure to get speed — Google by renting compute from SpaceX’s off-grid, gas-fired sites; Amazon via a permitted off-grid gas project near its Fayette County, Ohio data center — and the eventual winner depends partly on whether the current political permissiveness toward natural gas holds.

Amazon’s chief executive has called power the “single biggest constraint” in its cloud and AI business, which reframes the hyperscaler race as a contest over electricity procurement rather than chips or models alone. Amazon leads today on installed scale, but the relevant question the source poses is forward-looking: who adds power fastest, and at what cost in money, time, reliability, and emissions. Because the existing grid and equipment supply chain cannot absorb every operator’s demand at once, each company must pick its priorities — and Amazon and Google have made opposite bets inside the same physical constraint, which is why their projected trajectories converge even as their methods diverge.

The Constraint That Reorders the Race

The analysis turns on one stated bottleneck. Amazon’s chief executive has said power is the “single biggest constraint” in its cloud and AI business — meaning the binding limit is not demand for compute but the ability to energize it. That single claim is load-bearing for everything else: it converts a story about technology leadership into a story about who can acquire gigawatts, on what timeline, and under whose permits.

Amazon enters with an incumbent advantage. As “the world’s largest cloud provider,” it has “been building a lot of data centers over the past two decades,” and Sergio Toro, founder of Aterio, credited it with a “really good understanding of building capacity” and a “methodical approach,” reinforced by “longstanding relationships with utilities and suppliers of scarce power equipment.” Amazon was also early — it “announc[ed] the first power purchase agreement with an existing nuclear power plant,” after which Microsoft, Google, and Meta signed their own deals with mothballed or soon-to-close nuclear plants. The incumbency is thus partly a head start in relationships and procurement, not only in megawatts installed.

Same Grid, Opposite Priorities

The source frames the choice explicitly: “hyperscalers have to pick their priorities: speed, cost, reliability or environmental impact.” That framing is the analytical core, because it makes the two leaders’ divergence a deliberate trade-off rather than a difference in competence. Amazon’s strategy “seems to emphasize cost and reliability,” Google’s “appears focused on clean energy.”

The mechanism behind the cost difference is build-versus-lease. Amazon “plans to build out most of its own capacity,” while Google “is expected to rely more heavily on leases” — about a quarter of Google’s expected 2030 capacity, per Aterio. The source states the tension plainly: “Self-built can take longer but is the cheaper option over the long term.” So Amazon is trading speed for durable cost advantage and control; Google is trading some long-run cost for faster standup. Neither is strictly dominant — each optimizes a different criterion, which is precisely why a single “who wins” answer resists the available evidence.

How Google Converts “Clean” Into Speed

The counterintuitive move in the source is that Google’s environmental priority does not cost it speed — under specific conditions, it buys speed. Michael Thomas, founder of Cleanview, said Google is “doing everything it can” to avoid fossil-fuel-dependent data centers, and it is “the only hyperscaler with an in-house, renewables-focused developer,” having bought Intersect Power earlier this year. Google was also “the first tech giant to start studying how its data centers might use less power during hours when demand is high on the grid” — an attempt to make load itself flexible rather than fixed.

The speed comes from co-location. At least three planned Texas data centers “will be able to skip the long queue to connect to the grid because they are being built next to solar and wind projects,” because Texas market rules “allow faster grid connection if the data center is co-located with a new source of power.” In two of the three, Google builds the solar and wind itself through Intersect Power, and “in all three cases, the solar and wind capacity well exceeds that of the data center.” This is the distinctive engine of Google’s catch-up: it sidesteps the interconnection queue — the very constraint slowing everyone — by bringing its own generation, turning a regulatory rule into a timing advantage.

The Gas Hedge Both Leaders Accept

Despite the contrast, the source documents a convergence at the margin: both leaders accept off-grid natural gas to get speed now. The cleanest illustration is Google’s own compromise. “Access to speedy power means accepting near-term trade-offs, even for a green-conscious company”: Google “will be renting computing capacity from SpaceX, which built out its data centers in record time because it used off-grid, natural-gas-fired power equipment.” SpaceX’s subsidiaries “were sued earlier this year over air pollution concerns” from those turbines — so Google’s clean-energy posture is qualified by what it rents, not only what it builds.

Amazon’s gas exposure is documented more circumstantially. The source names Microsoft’s “20-year agreement with Chevron to power its AI data center in Texas with an off-grid, natural-gas-fired power plant,” and says “both Meta and Amazon have plans for such projects, according to data compiled by Cleanview.” Amazon “hasn’t publicly confirmed its involvement,” but its Fayette County, Ohio data center “is the only planned source of large power demand near a permitted off-grid, natural-gas-power project,” per Cleanview’s Thomas. That inference — proximity plus a permit — is weaker evidence than Microsoft’s signed contract, and the analysis should weight it accordingly: it is suggestive, not confirmed. Notably, “in most cases, hyperscalers eventually want these data centers to connect to the grid,” so the gas plants read as bridges, not endpoints.

What Could Reverse the Standings

The source is explicit that the lead is not locked: “The ultimate winner of the AI race will depend in part on where political winds shift — today’s environment is permissive of natural-gas power, but it might not stay that way forever.” This is the hard-to-reverse risk worth isolating. Both leaders have committed capital and timelines to a gas-permissive regime; a tightening of permitting or emissions enforcement would not gently nudge the plans but could strand the speed strategy that depends on off-grid turbines — and the SpaceX air-pollution suit shows that pressure already exists in the system.

The second open variable is technology. “All the hyperscalers have also placed bets on novel technology providers” — “small modular nuclear reactors, advanced geothermal and novel batteries to solar energy beamed in from space,” with SpaceX and Google “even eyeing data centers in space.” The source’s own caveat is that “some new energy technologies might reach a breakthrough faster than others,” which means today’s standings could be reordered by which of these matures first — an outcome the source explicitly declines to predict. The estimate that should update most on new evidence is therefore not Amazon’s current lead (well established) but the 2030 convergence (a projection contingent on permitting holding and breakthroughs not arriving asymmetrically).

Additional considerations

The forward-looking claims rest on a single estimator. The capacity figures, the “most capacity through 2030” and “fastest rate” projections, and the convergence forecast all come from Aterio, which “tracks company announcements, utility filings, building permits and satellite data” — credible inputs, but one vendor’s model, with no second source or stated error band in the article. The qualitative clean-energy reads on Google come from Cleanview. The source establishes Amazon’s present scale firmly and its future lead provisionally; it does not establish how reliably leased capacity (a quarter of Google’s 2030 mix) will materialize, nor what happens to either strategy if interconnection queues or gas permitting change before the data centers are energized. Those are the load-bearing uncertainties the projection rides on, and the article leaves them open.

Analytical techniques used in this piece

This analysis applies the methods below. Each links to a short, plain-English explainer you can read and reuse.

Bayesian Hypothesis Network
Updates the probabilities of competing hypotheses as evidence accumulates.
Multi-Criteria Decision Analysis
Scores competing options against several weighted criteria at once.
Wicked Futures
Explores a long-horizon, deeply entangled future with no clean resolution.
Bayesian Reasoning
Starting from base rates and updating beliefs proportionally as evidence arrives.
Mutually Assured Destruction
Deterrence by guaranteeing that any attack is suicidal for the attacker.