There’s a psychology experiment everybody knows. A kid, a marshmallow, and a choice: eat it now, or wait and get two. The children who waited did better later in life. The lesson we all took from it was that patience is character.

I’ve been thinking about that test while watching how the country is choosing to power AI data centers. And I keep landing somewhere uncomfortable: we’re not even getting the impatient option right.

The path of least resistance

Utilities and developers facing unprecedented new load are reaching for gas. Not because anyone ran the numbers and concluded it was optimal. Because it’s the option that requires no new thinking.

Gas comes in one box. One vendor, one fuel contract, one permitting path, one line item a CFO recognizes. You can put it in an integrated resource plan without explaining anything. Nobody gets fired for it.

That’s not a technology decision. It’s a decision to avoid making a decision.

Energy Innovation has a name for the belief underneath it: firm fixation. The outdated assumption that only always-on resources paired with major transmission builds can serve data centers. Their term, and it’s a good one, because it correctly identifies the problem as a belief rather than a constraint.

Except it isn’t even fast

Here’s what nobody checks before reaching for the shortcut.

GE Vernova closed Q2 2026 with 116 GW of gas equipment backlog and slot reservation agreements, up from 100 GW three months earlier, and expects at least 125 GW under contract by year-end. CEO Scott Strazik told analysts the company is now taking reservations for 2031 deliveries. Siemens Energy ended its fiscal third quarter with a 69 GW firm backlog and lead times of three years or more; CEO Christian Bruch put the current wait at about three years. Mitsubishi’s large-frame backlog hit 35 GW, with orders booked last quarter scheduled for delivery between 2028 and 2030.

Order a turbine today and you’re looking at three years at best. Reserve a slot at the leading edge and you’re waiting until 2031.

So the easy answer has a multi-year line in front of it. The thing chosen for speed isn’t fast. It’s just familiar, and we’ve confused those two words for so long that nobody bothers to check anymore.

It isn’t cheap either. Energy Innovation’s modeling puts the fossil-heavy path at an added $29.7 billion a year on customer bills by 2030. The clean path comes in $5.1 billion a year lower — about 17 percent cheaper. Add a fuel price spike like 2022 and the gap widens to $13.5 billion a year.

Somebody pays that. It isn’t the developer.

What Google actually did

In November 2025, Google announced a data center in Haskell County, Texas, co-located with a new solar and battery plant — the first industrial park to come out of its partnership with Intersect Power and TPG Rise Climate. That project, Quantum, is 640 MW of solar and 1.3 GWh of storage.

Per Google and Intersect’s June 2026 announcement, Quantum began operations that month while the Google data center it serves had only recently begun construction.

Read that again. The power was running before the building was finished.

That is the fact that should end the argument about speed. Not a model, not a projection. An operating asset that beat its own customer to the finish line while everyone else queued for turbines.

The part most people skip

Google’s energy parks still use gas.

The Meitner Energy Center in Gray and Roberts Counties integrates more than a gigawatt of wind, solar, and battery storage with on-site gas-fired generation for reliability firming. Google’s own framing, announced June 4, 2026: the majority of the site’s power comes from clean energy starting on day one, with a minority share of demand met by firming.

I think that’s the most important detail in the whole story, and it’s the one that gets left out of both sides of this argument.

Google didn’t prove you can skip gas. Google proved you can demote it.

Gas went from being the entire plant to being the last slice — the firming resource that covers the hours the clean assets can’t. Same fuel, radically different role, and a completely different emissions and cost profile as a result.

The unimaginative move was never using gas. It’s using gas as the whole answer when it should be the smallest part of one.

So what did $4.75 billion actually buy?

Google closed on its acquisition of Intersect Power in March 2026 — roughly $4.75 billion in a transaction that began in December 2025. As pv magazine reported, it was the first time a major tech firm bought a renewable developer outright rather than signing power purchase agreements. Intersect came with 2.2 GW of operating solar and 2.4 GWh of storage, and roughly 10.8 GW expected online or in development by 2028.

The obvious objection writes itself: sure, Google can do this. Google has $4.75 billion.

But look at what that money bought. Not a patent. Not a breakthrough chemistry. Not a technology nobody else can access. Solar modules, batteries, and gas backup are available to anyone willing to go get them.

What Google bought was the ability to assemble — people who know how to site generation next to load, get it permitted, stitch it together behind a single point of interconnection, and sequence the construction so the power shows up when the building does.

Google bought coordination. That’s it. That’s the whole moat.

The actual failure

Which means the thing standing between this industry and a better answer was never technology, and it was never really policy either.

It was that assembling four things is harder than buying one thing, and most organizations will choose the one thing every single time — even when the one thing takes years to arrive.

We built an entire industry narrative around firm power being the only serious option for serving large load, and then never went back to check whether it was true after the alternative got cheap. The equipment sat there, available, on the shelf. What was missing was anyone willing to do the work of putting it together.

That’s not a constraint. That’s a failure of imagination wearing a constraint’s clothing.

It has a name now

The thing Google assembled is starting to get called something: Bring Your Own Power. Clean generation brought to the system by the customers creating the demand. Its companion is Bring Your Own Flexibility — data centers shifting workloads, trimming operations, or leaning on on-site storage during stressed grid hours instead of demanding firm capacity around the clock.

The flexibility number is the one that should stop people cold. Duke University’s Nicholas Institute found the grid could absorb roughly 100 gigawatts of new data center load nationwide with no new peak generation at all, if those facilities curtail just half a percent of their annual electricity use during peak hours.

Half a percent. Not a technology gap. A willingness gap.

That research has been available since early 2025. It didn’t move much of anything until the approach had a name people could say in a rate case.

Back to the marshmallow

Here’s the thing about that famous experiment that most people never learned.

When researchers ran it again with a larger and more representative sample, most of the effect disappeared once they accounted for family background. The kids who grabbed the marshmallow weren’t short on willpower. They came from circumstances where the second marshmallow didn’t reliably show up. They were making a rational bet on a system that had already failed them.

The test measures trust, not character.

So I’ll grant the industry that much. Nobody is waiting patiently for a grid that takes a decade to plan and another decade to build. That’s not impatience. That’s arithmetic.

But that excuse only holds if the choice is really between waiting and grabbing.

Google didn’t wait. Google reached over and took both marshmallows.

The rest of the industry is still standing in line for one that shows up in 2029.

Sources

— END —

Want these insights in your inbox?

Once-a-month clean energy marketing analysis. No fluff.