Amazon holds the largest share of US data-center and power capacity among hyperscalers today, and new analysis reported by The Wall Street Journal on June 24, 2026 projects that advantage persisting through the end of the decade, though Google will have narrowed the gap considerably by 2030.
The story is not about technology. It is about land, utility agreements, and years of regulatory patience. Amazon has been building large-scale data centers for roughly two decades. That tenure compounds: long-term power purchase agreements signed years ago at prices that no longer exist, transmission interconnection slots that take years to secure, and cooling infrastructure built into land with permits already in hand. None of those assets can be replicated quickly, regardless of capital available today.
Google has clearly treated this as a priority investment and will close a meaningful portion of the gap with Amazon before 2030. That is not a minor achievement. Building out grid-scale power infrastructure against constrained utility capacity and stretched permitting pipelines requires sustained effort over many years, and Google has committed to it. Still, the projection puts Amazon in a different tier heading into the back half of the decade.
Why does this matter beyond bragging rights between two of the largest corporations on earth? Because electricity has become the actual binding constraint on AI scaling, not chip supply and not software. The major frontier labs all have access to sufficient GPU capacity at some level of cost. What they cannot always purchase is the reliable, high-density power delivery those GPUs require to run continuously. Substations, transmission capacity, and grid interconnection are infrastructure with multi-year lead times and finite supply in any given geography. The hyperscaler that controls the most contracted power wins the most AI workloads.
For builders outside the hyperscaler tier, this dynamic shapes vendor strategy directly. A startup or mid-scale AI company renting compute from AWS has implicit access to Amazon’s power footprint; the same logic applies to Google Cloud customers. The gap between the two is narrowing, but choosing a cloud provider today means implicitly betting on that provider’s ability to deliver capacity at scale through the end of your product’s early growth phase. If Amazon is projected to hold the larger share of incremental US capacity through 2030, that is a capacity-reliability argument, not just a pricing one.
For smaller labs that cannot afford to own data centers outright, the implications cut further. They are structurally dependent on whichever hyperscaler offers the most reliable capacity at the moment they need to scale training runs. Power availability is not typically visible in a cloud pricing sheet, but it surfaces as availability constraints, queuing delays, and regional limitations during high-demand periods. Labs that have already learned this lesson are the ones negotiating multi-year reserved capacity agreements now rather than relying on on-demand access.
The competitive question for the rest of this decade is whether any other large player, whether Microsoft, Oracle, or a sovereign-backed data center operator, can accumulate sufficient power commitments to join Amazon and Google as a genuine third option at scale. Until that happens, the energy infrastructure gap reinforces the same vendor concentration that already characterizes the AI compute market.
Teams currently making long-term infrastructure commitments should treat projected power capacity as a first-order input alongside pricing and model availability when selecting their primary cloud provider for 2027 and beyond.
Reported by The Wall Street Journal on June 24, 2026.
