Is it a Bubble?

It seems to be every other day a news article or social media media posts asks the question about the AI infrastructure expansion, all asking "Is it a bubble?"

It seems like you can't read an article or see a headline on a social media post without seeing the same question about the rapid expansion of AI infrastructure: Is it a bubble? 

The parallels to past tech cycles — runaway spending, massive facilities popping up seemingly overnight, and sky-high expectations — make the question feel almost inevitable. But to understand whether this is hype or reality, we need to look beyond headlines and toward hard data about what’s been announced, what’s built, and what’s actually needed to support AI demand.

One recent industry forecast projects that global active data-center IT capacity — which includes both legacy and AI workloads — could grow from roughly 24 GW today to nearly 147 GW by 2035, a six-fold increase as AI workloads proliferate. This equates to adding about 122 GW of new capacity worldwide over the next decade. That forecast reflects genuinely anticipated demand rather than speculative announcements. (ABI Research)

However, there’s a stark gap between announced development plans and real, operational capacity. Numerous industry observers have noted that planned projects often outpace what is physically deliverable due to constraints like power delivery, cooling infrastructure, water access, and permits. One commentary from industry tracking charts shows that announced capacity projections consistently run well above what has been completed, not because demand isn’t real — but because infrastructure bottlenecks slow down delivery. (LinkedIn)

This divergence between plans and delivery is exactly where the “bubble” narrative finds traction. It’s easy to compare press releases touting multi-billion-dollar data-center builds with the slower pace of actual construction and utility hookups. But unlike the classic speculative bubbles of the past — where the underlying service or customer demand was uncertain — real long-term demand for AI compute power is already embedded in enterprise digital transformation strategies and cloud provider roadmaps. This makes the backdrop fundamentally different: companies aren’t building data centers on a whim — they’re doing so to power workloads that are already scaling rapidly.

Another key difference from past tech booms is that AI infrastructure isn’t just about servers; it’s about physics. You can’t conjure hundreds of megawatts of power delivery, efficient cooling, or grid connections overnight. These constraints act as a natural governor on how much capacity actually gets built versus what’s announced — and that’s why we’re seeing announcements far outstrip completions in many regions. The “hype” is often less about demand and more about expectations about pace. (LinkedIn)

So is AI infrastructure a bubble? The data suggests that while announced projects may outnumber built capacity in the near term, the demand underpinning those announcements isn’t speculative — it’s tracking toward real, measurable adoption growth. The smarter lens isn’t bubble or no bubble — it’s which projects are grounded in realistic timelines and infrastructure feasibility, and which are simply aspirational press releases.

That’s the question that will separate hype from real expansion in the coming years.