Nokia CEO Justin Hotard’s argument is that the pace of AI data-center construction reflects supply limits more than a lack of customer interest. He says customers would probably build at twice the current pace if those limits eased—but that is a hypothetical assessment, not a construction forecast or proof that all planned capacity will be profitable.
8
10
Why Hotard thinks demand is still ahead of supply
Hotard points to shortages of memory chips and available power as constraints on how quickly customers can build. In his view, the industry is still early in deploying AI technology, so demand does not depend only on the release of more advanced models.
8
10
That distinction matters: companies can expand their use of existing AI models even if frontier-model development slows. Hotard’s argument, as reported by The Next Web, is that deployment of current technology could continue to drive infrastructure needs even without a new frontier model for three years.
7
What Nokia supplies to the AI buildout
Nokia is not providing the AI models in this argument. It supplies optical and IP networking technology used to connect AI data centers. Hotard has described data-center networking as part of Nokia’s role in the buildout, and Nokia’s reported results also point to demand for IP and optical connections as customers expand AI infrastructure.
2
3
What the “twice as fast” claim does—and doesn’t—show
Hotard’s statement is evidence of his view of customer appetite, not a measured estimate of how much capacity the market needs. Even if customers want to build faster, that alone does not establish that every project will earn a sufficient return or that long-term demand will match current expectations.
Those are separate concerns from the physical bottlenecks Hotard describes. Anthropic CEO Dario Amodei, for example, has warned that large compute commitments can be risky when they rely on uncertain revenue projections. That critique concerns investment economics; it does not directly disprove Hotard’s claim about current customer demand.
19
AI safety is another distinct question. Arguments for slowing frontier-model development focus on risks from developing more capable systems too quickly. Hotard’s demand case is narrower: deploying AI systems that already exist may require more infrastructure, even if the pace of frontier development changes.
7
17
The clearest reading, then, is not that overbuilding has been ruled out. Hotard is arguing that supply constraints are limiting construction today, while the commercial returns, long-term demand and appropriate pace of AI development remain open questions.