'Speed to Powerless': Wärtsilä Dissects AI's Power Problems

The AI boom is creating an unprecedented appetite for electricity, forcing energy infrastructure into the spotlight as one of the biggest challenges facing the data centre sector.

While billions of dollars are being invested in new AI campuses across the Americas, connecting those facilities to reliable power is proving far more difficult than building them. Lengthy grid connection queues, constrained transmission networks and rising demand for generation capacity are increasingly threatening project timelines.

Wärtsilä believes the industry is approaching a tipping point, arguing that developers must place long-term energy planning at the centre of their investment strategies if they are to avoid running out of power before their facilities come online.

The company describes this growing disconnect as "speed to powerless", where data centres are completed faster than the electricity infrastructure needed to support them.

Why grids are struggling to keep pace with the AI boom

Wärtsilä explores these issues in its report, Beyond the Grid: Building the Power System for AI in the Americas, drawing on data from Lawrence Berkeley National Laboratory, the International Energy Agency and national energy planning organisations.

The report concludes that electricity infrastructure is becoming one of the biggest constraints on AI investment across the region.

By the end of 2025, around 2,600GW of generation and energy storage projects in the US remained stuck in interconnection queues, creating uncertainty for data centre developments reliant on new grid connections.

Elsewhere, Brazil increased its forecast for data centre transmission demand in 2030 by 60% only three months after publishing its national energy plan, illustrating how quickly electricity requirements are changing.

"The race to build data centres across the Americas is extraordinary in its pace and scale. But power is not keeping up," says Risto Paldanius, VP Americas at Wärtsilä.

"Interconnection queues stretch years, transmission capacity is saturating in the corridors where data centres are clustering, and equipment lead times are pushing delivery towards 2030 and beyond.

"The risk is not that the market slows down – it is that it builds faster than it can reliably power.

"We call that speed to powerless, and the evidence suggests these pressures could converge into a serious constraint."

Beyond transmission capacity, the report highlights water availability as another growing consideration for energy infrastructure. Parts of the US and Mexico face increasing water stress, reducing the suitability of some conventional generation technologies that depend on significant cooling resources.

Is on-site generation the future for data centres?

Rather than relying exclusively on utility connections, Wärtsilä believes AI developers should diversify their electricity supply by combining grid power with large-scale on-site generation.

The company expects so-called macrogrids to become increasingly important as AI campuses expand beyond 100MW of peak demand.

Unlike conventional microgrids, these systems are designed to operate independently for extended periods while retaining the flexibility to connect to the wider electricity network when additional capacity becomes available.

According to Wärtsilä, this hybrid approach provides greater resilience against grid bottlenecks while allowing operators to respond more effectively to future increases in electricity demand.

Its modelling also compares reciprocating internal combustion engine (RICE) technology with aeroderivative gas turbines across a 20-year operating life.

Using the company's assumptions, RICE systems achieve a levelised cost of electricity of around US$86/MWh compared with approximately US$111/MWh for gas turbines. For a 1GW AI data centre, Wärtsilä estimates that difference could translate into annual savings of roughly US$178m.

The report also highlights operational advantages beyond economics.

According to Wärtsilä, reciprocating engines can maintain full rated output in temperatures ranging from –45°C to 45°C while requiring negligible process water through the use of closed-loop cooling systems, making them particularly well suited to regions where both high temperatures and water scarcity are becoming increasingly important considerations.

"This is not a binary choice between on-site generation and the grid," Risto explains.

"The most resilient data centres will combine modular on-site RICE generation with flexibility, including grid supply where available, clean power contracts and long-term planning for eventual interconnection.

"The companies planning for that hybrid approach now will be the ones still operating efficiently in 2035, and the ones that are not planning for it risk finding themselves powerless at exactly the moment the market expects them to deliver."

Challenges across the Americas

Although electricity supply is emerging as a common concern across the Americas, Wärtsilä says each market faces its own energy infrastructure challenges.

In the US, lengthy interconnection queues, transmission congestion and permitting delays remain major barriers, particularly in PJM and ERCOT, while demand is also expected to continue growing in California, Arizona and the Pacific Northwest.

Brazil faces pressure to expand its transmission network quickly enough to accommodate rising electricity demand, alongside proposed policy measures including the Special Tax Regime for Datacenter Services (REDATA).

In Mexico, developers are increasingly investigating self-consumption energy models as electricity network expansion struggles to match construction activity.

Meanwhile, Chile's nationwide blackout in February 2025 has sharpened attention on energy resilience and dependable generation capacity, while Argentina continues to offer attractive renewable resources and investment opportunities despite ongoing concerns around grid investment and policy certainty.

For Wärtsilä, the message is clear: the AI race will not be won solely by those who build the fastest, but by those who secure resilient, flexible and long-term energy strategies capable of supporting increasingly power-hungry digital infrastructure.