Amazon Leads in Computing Power Scale, Google Expands the Fastest
Taylor Wilson
Amazon's U.S. data centers now draw roughly 9 gigawatts — equal to North Dakota's entire generating capacity and the largest footprint among hyperscalers — but Aterio projects Google is expanding fastest and will sharply narrow the gap by 2030.
How big is the power gap among the four cloud giants?
Amazon's self-built data centers consume about 9 GW. Microsoft and Google each use roughly 5 GW; Meta sits at about 4 GW.
In plain terms = Amazon alone burns nearly double what Microsoft or Google does — roughly one U.S. state's entire power output.
Amazon's CEO has publicly called electricity "the single biggest constraint" on its cloud and AI business.
This means → whoever secures more power faster can scale compute faster. Electricity has overtaken chips as the real bottleneck in the AI race.
Why is Google on track to catch up fastest?
Aterio — tracking filings, building permits, and satellite data — projects Amazon will still add the most absolute capacity by 2030, but Google will post the fastest growth rate.
Factor in third-party leased capacity, and Google narrows the gap with Amazon significantly.
This means → Google is running a hybrid playbook — build some, lease a lot. Aterio estimates roughly one-quarter of its 2030 capacity will be leased. Leasing is faster to deploy but costlier over time.
Where exactly do Amazon's and Google's strategies diverge?
Amazon prioritizes cost and reliability: long-standing relationships with utilities and scarce power-equipment suppliers give it a systematic build-out machine, mostly self-built.
Amazon was also the first tech company to sign a power-purchase agreement with an operating nuclear plant.
Google bets on clean energy: it is the only hyperscaler with an in-house renewables developer — it acquired Intersect Power earlier this year.
This reflects two very different wagers: Amazon bets "secure cheap, reliable power first"; Google bets "green power is the long-term moat."
What is Google doing differently in Texas?
Google has at least three planned Texas data centers that can bypass the lengthy grid-interconnection queue — they sit adjacent to solar and wind project sites.
Two of the three will have renewable capacity built by an Intersect Power subsidiary, and all three projects have renewable installations that far exceed the data centers' own power needs.
This means → Google is not waiting for grid approval. It is generating its own power on-site — a "power follows the data center" approach that buys time.
Does the "clean" route still require compromise?
Google will lease compute from SpaceX, whose data centers were built quickly using off-grid natural-gas generators — a SpaceX subsidiary has already been sued this year over air pollution from those turbines.
Microsoft went further: a 20-year deal with Chevron to power its Texas AI data center with an off-grid gas plant.
Meta and Amazon have similar off-grid gas plans. Amazon has not publicly confirmed its project, but its Fayette County, Ohio data center is the only plausible large-scale customer for a nearby permitted off-grid gas facility.
In plain terms = everyone talks green, but when speed pressure mounts, natural gas becomes the universal "bridge solution."
What will ultimately decide this race?
Amazon builds mostly in-house — longer lead times, but lower long-term costs.
Google leans on leasing — faster to come online, but pricier and carrying a simultaneous clean-energy pledge.
This means → whether Google can maintain the fastest expansion while honoring its green-power commitments is the critical test of its strategy's sustainability.
Content is for reference only, not financial advice.