AI Energy Infrastructure Investing: Which Power Assets Really Win

Power is supposed to be the easy part of the AI trade, yet roughly 220 GW of gas turbine backlog and reservations now sits across just three manufacturers. Developed-market electricity demand is growing for the first time in over two decades, and the scarce asset is no longer compute but speed to power.

That scarcity shapes AI energy infrastructure investing more than any single demand forecast. Hyperscaler capex forecasts run from hundreds of billions of dollars to well over $1 trillion a year, and the real question is which power assets capture that spend with returns that last.

Some will. Others carry merchant risk (selling electricity into wholesale grids at fluctuating prices) that the headline numbers hide.

Here is a framework for judging which power assets benefit from AI demand, and which ones only look like they do.

Why is electricity demand rising again, and how much of it is really AI?

Global electricity demand grew about 4% in 2024, or 1,172 TWh, according to Ember’s Global Electricity Review 2025. Developed markets rebounded after two decades of flat or falling demand.

AI is only part of that story. Weather did more of the lifting than most headlines admit.

The growth drivers

Heatwaves and cooling added about 0.7 percentage points of the 2024 growth (208 TWh), per Ember. Data centres and cryptocurrency together added about 0.4 percentage points, alongside electrification of transport and heating.

Because weather and electrification contribute comparably to AI, you should treat AI demand as one durable driver among several. Underwriting a power asset on an AI-only thesis leans on the least certain slice of the growth.

The International Energy Agency (IEA) data points also differ by vintage. Its estimates run about 415 TWh for 2024 and 485 TWh for 2025, with an updated central case near 950 TWh by 2030, while an earlier base case ran higher at more than 1,000 TWh.

Forecast ranges differ widely because AI data center energy demand depends on workload growth, chip efficiency and cooling design, which is why investors should stress-test any single-point estimate before underwriting a power asset against it.

Supply matters as much as volume. The IEA’s mix for data-centre electricity is:

• Coal: about 30%
• Renewables: about 27%
• Gas: about 26%
• Nuclear: about 15%

Regional snapshots: Australia and Southeast Asia

Indonesia’s data-centre use is expected to quadruple from 6.7 TWh in 2024 to 26 TWh by 2030, according to Ember’s ASEAN analysis. In Australia, an EQT speaker says data centres’ share of grid capacity will climb from 3% to 13%, while east coast demand rises from about 170 TWh to 400 TWh by 2040.

Local load growth is rapid in both. EQT also argues that 75% of the global population faces energy security risk, a point that returns later.

Why do investors pay for contracted, on-site power and avoid merchant exposure?

Start with the mechanics. A merchant plant sells electricity into the wholesale grid, so its revenue rises and falls with spot prices.

A contracted asset sells under a long take-or-pay agreement, where the buyer pays for the capacity whether or not it uses all of it. When that buyer is a high-credit hyperscaler, the cash flow looks more like a toll road than a commodity business.

The 15-20 year take-or-pay structure EQT targets businesses with 15-20 year take-or-pay contracts, which lock in revenue regardless of short-term power prices.

Speed explains the appeal of on-site supply. Data centres take roughly 1-3 years to build, often faster than grid generation and transmission can be delivered. Large hyperscale sites commonly draw 100 MW or more and need firm, around-the-clock power, though that figure is lightly verified.

EQT, the Swedish alternative asset manager with Wallenberg family roots and a long-term industrial approach, offers a practical screen:

1. Speed to power.
2. Long-term contracted revenue, typically 15-20 years.
3. Proven development track records, which allow earlier offtake contracting.
4. Caution on undifferentiated merchant exposure.

Scale Microgrids, a US-based EQT portfolio business, shows the model in practice. It builds on-site solar, batteries and gas for data centres and heavy industry in the US and Europe, though no financial metrics are available.

When you evaluate any power investment tied to AI, contract structure and counterparty credit say more about downside protection than the headline demand forecast does.

Where is the capital going: gas turbines, solar, batteries and the shrinking coal share?

Follow the scarcity. The tightest bottleneck is gas turbines, and the numbers show it quickly.

The gas bottleneck

EQT cites backlogs of 116 GW at GE Vernova, 69 GW at Siemens Energy and 35 GW at Mitsubishi, roughly 220 GW in total. GE Vernova expects to be mostly sold out through 2030; earlier, lower figures reflect different dates and definitions, so the recent number is preferred.

Sold-out slots hand pricing power to manufacturers and slot holders. You should look for exposure to scarce delivery positions rather than assume new gas capacity can be added quickly.

The demand pulling on that capacity is large. Goldman Sachs forecasts, cited by Nick Griffin, put hyperscaler capex at:

• About $800 billion in 2026
• $1.2 trillion in 2027
• $1.4 trillion in 2028

EQT expects capex in its focus subsectors to grow from about $100 billion to about $425 billion, roughly 20% a year.

Hyperscaler spending of this scale is forcing a rethink of global power infrastructure, from turbine supply chains to transmission upgrades, and the pace of delivery will decide which assets capture the capex.

Solar, batteries and the coal retreat

Solar and batteries win on speed and cost. Solar added a record 600 TWh of generation, and storage is the fastest-growing power technology, though no quantified global battery deployment figure was available.

Coal is losing ground. It now supplies under one-third of energy demand for the first time, hit by US policy volatility and ageing plants, while renewables reach about 32% of global electricity. A hybrid of gas, solar and storage is often the fastest route to firm power.

What could go wrong, and will sustainability become security?

The bullish case is clear. Demand is rising, supply is tight and long contracts are being signed.

Nick Griffin, whose growth-equity view frames AI as an infrastructure boom rather than a bubble, says the build-out is about three years into a decade-long run. He estimates under 5% of eventual AI usage has occurred and points to $60 billion of incremental quarterly hyperscaler revenue, though these are his views rather than verified data.

A boom that will eventually end Griffin compares the cycle to the Australian resources super cycle and acknowledges the boom will eventually end.

Where the thesis is fragile

Several risks cut against the story:

• Forecasts are scenario-dependent and sensitive to workload growth and efficiency gains.
• A possible AI overbuild could leave capacity idle.
• Merchant projects face cyclicality, and interconnection queues often exceed three years.
• High-emissions assets risk being stranded if policy tightens.
• Efficiency standards, renewable procurement rules and carbon pricing could change project economics.

Ember and Reuters stress the weather component of recent growth, which tempers the AI narrative. No public statements from Brookfield, KKR or BlackRock/GIP on AI-driven on-site power were found.

You should size positions on the assumption that part of the current build-out may prove cyclical, favouring contracted, high-credit exposure that survives a demand pause.

From sustainability to sovereignty

The theme itself may be shifting. Turbine scarcity, the premium on firm capacity and the push for dedicated local supply point towards energy security, with EQT’s speaker expecting sustainability trends to evolve into sovereignty trends.

That framing could extend the demand story beyond AI. It also means a theme already priced in as “green” may be repriced as “secure”.

For readers weighing security-driven themes, our deep-dive into geopolitical energy investment frameworks shows how to position portfolios when supply shocks reshape capital flows.

Backing speed, contracts and scarcity over headline demand

Demand growth is real, but its AI share is only partly structural. Three filters hold up: contracted revenue with high-credit counterparties, speed to power, and exposure to scarce delivery positions.

Watch four things next: turbine delivery schedules, contract terms in new deals, interconnection progress and any softening in hyperscaler capex. Apply the screen to your own energy holdings and see which ones pass.

This article is for informational purposes only and should not be considered financial advice. Investors should conduct their own research and consult with financial professionals before making investment decisions. Past performance does not guarantee future results. Financial projections are subject to market conditions and various risk factors. These statements are speculative and subject to change based on market developments and company performance.

—