Nokia's CEO Says Data Centers Would Go Up Twice as Fast if Supply Allowed It — and the Constraint Isn't Demand

Every few weeks someone declares that the AI data center boom has gone too far. Nokia's chief executive Justin Hotard has a different objection to the whole framing: it assumes demand is the variable that matters. Speaking on CNBC's The Tech Download, he argued the opposite — that the industry is not overbuilding, it is underbuilding, and the proof is in the supply chain.

"I don't think you can say we're overbuilding at all, because the reality is — if we could build 2x faster, our customers could build 2x faster, they probably would," Hotard said. He named two shortages doing the limiting: memory chips and energy. And he pushed back on the idea that demand depends on the next model release. "Even if we didn't have another frontier model released in the next three years, we could probably make tremendous progress just deploying the technology that's there today."

Hotard is not a neutral observer. He ran Intel's data center and AI group before Nokia hired him in 2025, and Nokia sells the connective tissue of the buildout — the equipment that links racks inside a data center and data centers to each other. That business is doing well: data-center connectivity sales doubled to €446 million in the second quarter, against group sales of €4.82 billion, and Nokia's shares are up roughly 130% over the past year.

The memory squeeze he describes is visible well outside the data center. DDR5 prices in Germany have risen 414% in a year, and memory can now account for as much as half the bill of materials on a budget smartphone. That is a strange kind of shortage — one that shows up in a phone made in Shenzhen as sharply as it does in a training cluster in Virginia.

The financial counter-argument is serious, and Hotard does not really answer it. Bain said last week that AI needs \$6 trillion a year in revenue by 2031 to pay for the data centers now under construction, against the \$1.2 trillion to \$1.8 trillion today's AI products could actually generate. A paper by Stijn Van Nieuwerburgh presented at the Brookings Papers on Economic Activity puts the investment bill at \$10.3 trillion between 2025 and 2032 — an average of 3.63% of American GDP a year, larger relative to the economy than the canal, railway, electrification, highway and telecommunications booms. He also warns the financing is migrating off balance sheets into joint ventures, private credit and special purpose vehicles, which makes the true leverage much harder to see.

Europe, where Nokia is headquartered, illustrates the gap between ambition and money. The EU opened bidding in July for seven AI gigafactories worth €30 billion; only about €1 billion of public money is actually committed. Most of Nokia's customers are not in Europe, which is precisely why the company can grow its connectivity business while its home continent holds a debate.

Hotard's argument is self-serving, but it is also testable, which is more than can be said for most of the AI-capex debate. If he is right, memory pricing and power interconnect queues should stay tight even as model releases slow, and buildout should accelerate the moment supply eases. If the buildout slows anyway — if customers start walking away from leases and GPU prices fall while memory is still expensive — then demand was the constraint all along and the \$10.3 trillion question becomes urgent very quickly. Watch the memory market first. It is the constraint that shows up in your phone before it shows up in a data center.