The AI boom isn't slowing down. It's tripping over itself. Speaking on the All-In Podcast during Nvidia Corp‘s (NASDAQ:NVDA) GTC, IREN Limited (NASDAQ:IREN) CEO Daniel Roberts described the current AI infrastructure buildout in blunt terms: "permanent whack-a-mole."

That may be the most accurate description of the industry right now.

From GPUs To Power — And Now Time

The constraints keep shifting. First, it was GPUs (graphics processing unit). Then it became power. Now, according to Roberts, the real bottleneck is something else entirely: "time to compute."

Even as companies secure chips and energy, they still can't deploy fast enough. Why? Because AI isn't just software—it's physical.

Building data centers means: securing land, connecting to the grid, hiring thousands of workers and constructing cooling and electrical systems.

And that's where the system starts to break.

The Real World Can't Keep Up

Roberts put it plainly.

The "real world" is struggling to keep up with "digital exponential demand curves."

Every time one constraint is solved, another emerges:

  • fix chip supply → power becomes scarce
  • solve power → labor becomes the issue
  • solve labor → supply chains tighten

That's the whack-a-mole.

And it's constant.

No Idle GPUs, No Slowdown

If anything, demand is accelerating.

"There are no idle GPUs in the world," Roberts said, adding that the industry still "cannot meet demand."

That's a critical signal.

This isn't a bubble struggling to find use cases—it's an ecosystem struggling to keep up with them.

The Bigger AI Picture

The AI trade is evolving. It's no longer just about chips or models. It's about infrastructure—and the limits of building it fast enough.

And if Roberts is right, the next phase of AI won't be defined by breakthroughs in software.

It'll be defined by how quickly the real world can catch up.

Image: Shutterstock