Artificial intelligence has attracted enormous capital. But as spending increases, investors and lenders are becoming less interested in what an AI project could become and more interested in whether it can actually work economically.

JPMorgan Co-head of Global Investment Banking Kevin Foley described this on CNBC’s “Squawk Box” on Sept. 1. 

The money is there, but the scrutiny has gotten tougher. And of course, one factor stands above the rest when it comes to making or breaking a financing decision. Power.

Obviously you're going to care about power sources and permits.

Foley said these factors are crucial in separating projects that get funded from those that don’t. He added the track record of developers and the nature of partnerships as additional factors, but power came first because it appears to be most important.

Also Read: GE Vernova Latest News and Stories

What JPMorgan is actually seeing in the AI financing market

Foley pushed back on the narrative that capital is pulling back from AI infrastructure, according to CNBC. 

“We have not seen any projects getting slowed down because of a lack of financing,” he said. The pipeline is active. But the market has become, in his words, “very discerning.”

That discernment has an attached cost. More supply of capital chasing fewer high-quality projects pushes rates higher. Foley framed it as “capital markets 101” — straightforward economics applied to an asset class that briefly seemed immune to normal rules.

More Energy:

  • Goldman Sachs sends surprising message on oil prices
  • BofA commodities expert reveals what must happen before oil prices fall
  • Chevron CEO evaluates new pipeline route around the Strait of Hormuz

The scale of what’s being financed helps explain why the scrutiny is rising. JPMorgan’s own Q2 2026 earnings call data proves it. Global AI-specific capital expenditure grew from $400 billion last year to $700 billion this year, and it’s projected to surpass $1 trillion in 2027.

Goldman Sachs’ August report showed that estimated total global AI-related investment will reach $1 trillion in 2026 alone, and $1.8 trillion in estimated cumulative global investment in AI since 2022 by the end of 2026.

At that scale, the difference between a well-structured project and a poorly structured one is hundreds of billions of dollars in capital allocation that either compounds or gets written down.

Why power is the filter JPMorgan catches first

Foley’s mention of power sourcing is something that anyone building or financing AI data centers has faced. Getting electricity to a new facility, quickly and reliably, is often harder than building the facility itself.

The International Energy Agency (IEA) projects that global electricity generation serving data centers will rise from approximately 460 terawatt-hours (TWh) in 2024 to more than 1,000 by 2030 in its base case, the IEA reported.

Data center consumption is expected to expand from 1% of total global electricity generation today to 3% by 2030. Under a high-growth scenario, consumption could approach 2,000 TWh by 2035.

That’s a huge amount of new power demand arriving in a short window. And it links us directly to the constraint Foley flagged: If you can’t demonstrate a credible power source and the permits to use it, the project doesn’t get financed. Full stop.

GE Vernova supplies roughly 25% of the world’s electricity through its equipment.

Kyle Grillot/Bloomberg via Getty Images

Why GE Vernova sits directly on the bottleneck JPMorgan is describing

This is where GE Vernova (GEV) enters the chat. How? Not as an incidental beneficiary of AI enthusiasm, but as a direct provider of solutions to the very constraint that could make or break project financing decisions.

GE Vernova supplies roughly 25% of the world’s electricity through its equipment, according to its statement. Its gas turbines, grid infrastructure, and electrification systems are exactly what data center developers need to satisfy the demand.

I found out that GE’s own data center momentum is accelerating — aggressively, in ways the headline revenue numbers don’t fully capture. 

Related: Morgan Stanley points to the good news in Marvell’s data centers

Data center orders for electrification equipment totaled more than $5 billion year to date in 2026. It’s actually more than double the total for all of 2025, according to the Q2 2026 earnings report. 

In fact, GE Vernova confirmed that its Electrification segment booked $2.4 billion in data center equipment orders in Q1 2026, more than all of 2025.

Management noted that customers want power that is “reliable, scalable and efficient — and they want it quickly.” That’s the description of the JPMorgan financing filter to real product demand.

GE Vernova’s Q2 fiscal 26 earnings justify that orders are already responding

  • Orders reached $24.2 billion, up 88% organically. 
  • Gas Power equipment backlog and slot reservations grew from 100 gigawatts to 116 gigawatts, with expectations to reach at least 125 gigawatts by year-end.
  • Free cash flow hit $5.1 billion in Q2 alone,  more than all of 2025.
    • Source: GE Vernova Second Quarter 2026

My read on the connection is that Foley’s comment on power sourcing being a primary filter for AI project financing signals where durable value could be building. 

Yes, the companies building the models and chips get most of the attention. But the ones keeping the power on could also quietly become some of the market’s most important infrastructure plays.

Related: Morgan Stanley strongly resets GE Vernova stock target