Interconnection queues and grid limits could strand projects across the Americas
Data centre developers across the Americas are building faster than power systems can supply them, and a new whitepaper from technology group Wärtsilä warns the gap could become a material brake on AI infrastructure growth across the region.
The analysis, Beyond the Grid: Building the Power System for AI in the Americas, points to interconnection queues, transmission constraints, long equipment lead times and permitting delays converging across the United States, Brazil, Mexico, Chile and Argentina.
Wärtsilä calls the risk "speed to powerless," where projects secure approvals but cannot secure the power to operate.
Risto Paldanius, vice-president, Americas at Wärtsilä, said the risk is not that the market slows down but that it "builds faster than it can reliably power."
That gap is what the company calls "speed to powerless," he said.
The scale of the backlog is clearest in the US, where about 2,600 GW of generation and storage capacity was seeking grid interconnection at the end of 2025, according to Lawrence Berkeley National Laboratory data cited in the report.
For developers relying on a traditional grid connection, the company argues, that queue creates serious uncertainty over both the timing and the cost of firm supply.
Demand assumptions are also shifting quickly.
In Brazil, projected data centre load on the transmission grid for 2030 was revised upward by 60 percent, from 18.9 TWh to 30.3 TWh a year, just three months after the country published its Annual Energy Operation Planning for 2026 to 2030,
Signed grid-connection contracts rose from 8 to 22 over the same window.
Wärtsilä argues the response lies in what it defines as macro-grids, fully or partially isolated systems drawing more than 100 MW of on-site peak load.
The company positions these as a step beyond smaller microgrids and a bridge to eventual grid connection rather than a permanent replacement for it.
Much of the report makes the case for the company's own generation technology.
Its 20-year modelling compares reciprocating internal combustion engine (RICE) systems with aeroderivative gas turbines and, under the assumptions used, puts the levelized cost of electricity for RICE about 25 percent lower, at roughly US$86/MWh against US$111/MWh.
Wärtsilä translates that into annual savings of about US$178m for a one-gigawatt facility.
The analysis also credits reciprocating engines with holding full rated output from -45°C to 45°C, while aeroderivative turbines can lose up to 27 percent of output at high ambient temperatures, and with drawing negligible process water, a point the company flags for water-stressed hubs such as Querétaro in Mexico.
The report frames each market individually.
In the US, interconnection delay, local transmission congestion and permitting friction are the principal barriers, concentrated in the PJM and ERCOT regions, with further growth expected in California, Arizona and the Pacific Northwest.
In Brazil, centrally planned transmission expansion means delivery capacity to data centre corridors can lag by years, while the proposed Redata tax regime, under Senate review as of May 2026, is shaping technology selection.
In Mexico, the mismatch between fast construction schedules and slower grid expansion is driving interest in self-consumption models.
In Chile, corridor saturation and the February 2025 national blackout have sharpened attention on resilience.
In Argentina, strong wind and gas resources and the Large Investment Incentive Regime sit against chronic grid underinvestment and policy uncertainty.
The whitepaper argues power procurement is now an investment-certainty question as much as an engineering one, noting that on-site assets can be repurposed or monetized in energy and capacity markets once a facility connects to the grid.
"This is not a binary choice between on-site generation and the grid," Paldanius said.
The most resilient facilities, he argued, will pair modular on-site generation with grid supply, clean power contracts and long-term planning for interconnection.
