Speed to Powerless”: The Power Gap Closing In on Americas AI Build: the real signal is the immediate adjustment required in cash, risk, and execution
Signals That Are Accumulating
The evidence is not coming from a single market or a single quarter. It is arriving simultaneously from multiple jurisdictions, and the direction is consistent.
In the United States, approximately 2,600 GW of generation and storage capacity were seeking grid interconnection as of the end of 2025, according to analysis published by Wärtsilä drawing on data from Lawrence Berkeley National Laboratory and the International Energy Agency. That figure does not represent future demand—it represents projects already in queue, competing for interconnection capacity that transmission infrastructure cannot absorb at that rate. The timing uncertainty created by that backlog is structural, not cyclical.
In Brazil, the Annual Energy Operation Planning for 2026–2030 revised its projected data center load on the transmission grid upward by 60 percent—from 18.9 TWh to 30.3 TWh per year—within three months of publication. A revision of that magnitude in that timeframe is not a forecasting error; it is evidence that market assumptions are being overtaken by demand faster than any planning body can formally track.
In Mexico, water scarcity is eliminating technology options before grid access even becomes the question. Aeroderivative gas turbines—a standard choice for large on-site generation—carry significant process-water requirements for cooling, which directly narrows viable power architectures in hubs like Querétaro where water supply is already constrained. Chile experienced a national blackout in February 2025 that has since concentrated attention on firm capacity and resilience requirements in a market where corridor-level transmission saturation was already present. In Argentina, strong natural resources exist alongside grid underinvestment and policy instability that make execution risk difficult to price.
No individual market is an anomaly. The pattern is regional.
Why No One Is Naming It Yet
The reason this convergence has not produced a visible industry response proportionate to its scale is structural: data center development cycles and power infrastructure cycles do not share a planning horizon.
A hyperscale data center can be sited, permitted, and under construction within eighteen to thirty-six months in a favorable market. Transmission expansion, interconnection queue resolution, and large power transformer procurement operate on timelines of two to seven years. The gap between those two rhythms is where Wärtsilä’s analysis identifies what it calls “speed to powerless”—the condition in which projects are built faster than the power systems required to run them can be delivered.
A secondary reason is that the signals are fragmented across jurisdictions. The Brazil revision looks like a local planning anomaly. The US interconnection queue looks like a regulatory backlog awaiting reform. The Mexico water constraint looks like a site-selection issue. Viewed individually, each is a regional problem with a regional solution. Viewed together, they constitute a systemic constraint on the AI infrastructure layer that every major operator is building simultaneously. That systemic read requires crossing market and regulatory silos that most teams do not routinely bridge.
It is worth noting that the analysis framing this pattern was published by Wärtsilä, a company with a commercial interest in the power technology conclusions it reaches. The underlying data sources—Lawrence Berkeley National Laboratory, the IEA, and national energy planning bodies across the Americas—carry independent weight, but the architectural recommendations in the analysis should be read with that commercial context in mind.
What Happens If the Pattern Continues
If interconnection queues remain saturated and equipment lead times hold at current levels, competitive differentiation between data center operators will increasingly be determined by power decisions made in 2025 and 2026—not by infrastructure decisions made closer to operational need.
The financial exposure is material. Wärtsilä’s analysis models a 20-year levelized cost comparison between reciprocating internal combustion engine systems and aeroderivative gas turbines for large on-site generation. Under its modelled assumptions, the RICE configuration carries a 25 percent LCOE advantage—approximately $86/MWh versus $111/MWh—translating to roughly $178 million in annual savings for a 1 GW facility. Those figures originate from a vendor-commissioned model and are not independently audited. But even directionally, LCOE differentials of that magnitude over a 20-year asset life represent a strategic consideration that belongs in procurement planning now, not at final investment decision.
The stranded capacity risk runs in both directions. Operators who build without securing firm power face operational delay or curtailment. Operators who secure generation capacity without a credible path to eventual grid integration may find themselves holding high-cost isolated infrastructure as grid expansion eventually reaches their corridors.
What You Can Do Before It Is Obvious
The window in which on-site power architecture decisions carry the most option value is the period before interconnection queue resolution and equipment lead times force a default choice. Based on current queue depths and transformer procurement timelines, that window appears to be closing through 2026 and 2027 in the markets where AI infrastructure density is highest.
Three moves have more value now than they will in eighteen months. First, map your portfolio against transmission corridor saturation in PJM, ERCOT, and the Brazil and Chile data center corridors—not at the site level but at the regional corridor level, where the constraint actually sits. Second, model technology selection for new-build capacity against water availability as a co-equal variable alongside interconnection timing; this is already a decision driver in Mexico and Arizona and will extend further. Third, pressure-test whether your current PPA and interconnection strategy carries an implicit assumption of grid availability timelines that the queue data no longer supports.
Operators who treat on-site generation as a bridge architecture rather than a permanent substitute—maintaining optionality for eventual grid integration while securing firm capacity now—are building the most durable commercial position. That is not a vendor conclusion. It is the operational logic of planning for infrastructure that needs to perform across a 15-to-20-year asset life, not just at commissioning.
Sources
- Datacenterdynamics — The Americas’ data center boom may soon face a critical power crunch, Wärtsilä warns – and the industry is (Link)
