The AI Data Center Boom: Massive spending, hidden cracks, and what doesn’t add up

The AI data center boom is bigger than it looks

The scale of the AI data center boom is hard to fully grasp until you put real numbers next to it.

In 2025 alone, major tech companies reportedly set aside around $400 billion for AI infrastructure. That’s not a slow build. That’s a full-speed, all-in commitment to a future powered by data centers, chips, and compute.

What’s striking isn’t just the number, it’s what it represents. More money is now going into building and equipping data centers than into housing in some parts of the economy. And that figure doesn’t even include the hidden costs: energy, staffing, maintenance, or the billions flowing through private companies.

On paper, it looks like unstoppable momentum.

But when you step back, the picture starts to feel less stable.

A boom without profits

For all the excitement around AI, there’s an uncomfortable truth sitting underneath it: Most companies still haven’t figured out how to make consistent profits from it.

Yes, the technology is powerful. Yes, adoption is growing. But turning that into sustainable revenue at scale is still a work in progress.

The only clear winners so far have been the companies selling the infrastructure, especially chip manufacturers. They don’t need AI to succeed commercially. They just need demand to keep flowing.

And right now, demand is everywhere. Or at least, that’s what the narrative suggests.

The contradiction at the center of the boom

Here’s where things get difficult to ignore.

At the same time companies are announcing massive investments in new data centers, a large portion of those projects are being delayed or quietly scaled back. Some never move beyond early-stage construction.

That creates a strange contradiction. If demand is truly exploding, why isn’t the physical infrastructure keeping up?

Part of the answer lies in how these projects are presented. Announced capacity often sounds impressive, but “under construction” can mean anything from a nearly finished facility to a site where only the foundation has been poured.

In reality, the gap between what’s promised and what’s operational is much wider than it appears.

Where are all the chips going?

This question becomes even more important when you look at the supply side.

Companies like Nvidia are producing massive volumes of AI chips, with output measured in gigawatts of computing power. But global data center capacity, at least what’s verifiably active, doesn’t seem large enough to absorb all of it.

Even if you assume rapid expansion, there’s still a mismatch. Not all the power in a data center goes to computation. A significant portion is used for cooling, networking, and storage.

So the real question isn’t just about production. It’s about utilization.

If the infrastructure isn’t ready, then a portion of that hardware isn’t being used at full capacity, or at all.

The real bottleneck isn’t what you think

It’s easy to assume that advanced chips are the limiting factor in the AI data center boom.

But increasingly, they’re not.

The real constraint is power.

Building a data center is one thing. Powering it is another. Electrical infrastructure: transformers, grid connections, energy supply has become one of the biggest bottlenecks in the entire system.

In some cases, fully built facilities are sitting idle, waiting for the grid to catch up. In others, projects are delayed simply because the energy requirements can’t be met yet.

It’s a reminder that no matter how advanced the technology becomes, it still depends on very physical, very finite resources.

Why spending keeps accelerating anyway

Despite all these constraints, companies aren’t slowing down. If anything, they’re doubling down.

The reason is simple: uncertainty.

If you hesitate, you risk falling behind. If you wait, you might not get access to the next generation of hardware. And in a market moving this fast, that risk feels bigger than overpaying or overbuilding (does this means we live in the AI buble?)

So companies buy early. They secure supply before they need it. They invest ahead of demand. This creates a feedback loop where spending increases faster than actual usage, a classic setup for inefficiency.

The warning signs are starting to show

In a market where demand is supposedly insatiable, rising inventories don’t quite make sense. Yet some suppliers are reporting exactly that, more hardware sitting on balance sheets than before.

It doesn’t mean the boom is over. But it does suggest that something isn’t perfectly aligned.

At the same time, energy costs are climbing. Running these data centers is becoming more expensive, not less. And as those costs rise, the margin for error shrinks. Suddenly, decisions that made sense in a low-cost environment start to look riskier.

The problem with moving too fast

There’s another issue that doesn’t get talked about enough: time.

AI hardware evolves quickly. What’s cutting-edge today might feel outdated in a year or two. But infrastructure investments are long-term by nature.

That creates a tension.

Companies are spending billions on hardware that may lose relevance faster than expected. At the same time, they’re accounting for those investments over much longer periods. The result is a system that looks stable on paper but may be more fragile in practice.

So where does this leave the AI data center boom?

The AI data center boom isn’t a myth. The spending is real. The ambition is real.

But so are the constraints. Power limitations, delayed projects, rising costs, and uncertain returns are all part of the same story. And right now, they’re being overshadowed by momentum.

That doesn’t mean the boom will collapse. But it does mean it’s more complex than it first appears.

Final thoughts

Every major technological shift comes with a phase where investment runs ahead of reality.

The AI data center boom feels like one of those moments.

There’s enormous potential, but also growing pressure beneath the surface. The next phase won’t just be about building more, it will be about making what’s already built actually work.

Sources:

International Energy Agency (IEA) - Data Centres and Energy https://www.iea.org/reports/data-centres-and-data-transmission-networks

Goldman Sachs - AI and Data Center Investment Outlook https://www.goldmansachs.com/insights/articles/data-centers-could-boost-european-power-demand-by-30-percent

Bloomberg – Global Data Center Expansion https://www.bloomberg.com/news/newsletters/2026-04-12/oracle-hire-shows-cfos-growing-role-in-ai-boom