Nokia CEO Justin Hotard (저스틴 호타드) pushed back against the view that AI data centre investment is excessive, saying the industry could build data centres twice as fast as it does now if not for supply constraints.

On Oct. 5, CNBC reported that Hotard assessed the current market as still in an early phase of demand for AI infrastructure.

In a CNBC podcast interview, Hotard said it is hard to say the industry has entered a phase of overbuilding. "You cannot say today that it is overbuilding," he said. "If we could build 2 times faster and customers could build 2 times faster, we probably would have done so." A shortage of key memory chips and power constraints were cited as the main factors slowing the pace of data centre expansion.

The remarks came as debate continues across the AI industry over whether investment in data centres, semiconductors and power infrastructure is moving too fast. Concerns remain in the market that long-term demand for AI services must be sustained for large-scale investment to be justified. The outlet said investment in AI buildout, including data centres, chips and related infrastructure, was projected at $10.3 trillion from 2025 to 2032.

Nokia is shifting its focus from a traditional telecommunications equipment maker to an AI infrastructure company. It supplies network technology that connects chip racks inside data centres and technology that links data centres across regions. As it plays a so-called "picks and shovels" role in the AI boom, investors' assessments have also changed. Nokia shares have risen about 130 percent over the past year.

Market unease grew once last month. After warnings by Anthropic researchers, discussion emerged that development of advanced AI models should be slowed, and selling spread in AI-related stocks as Dario Amodei called for slowing development of the most advanced models.

Hotard, however, stressed that AI demand should not be judged only by whether frontier models are released. "Even now, if you look at the deployment phase, it is very early," he said. "Because these models are so powerful, just deploying them creates a lot of demand." He added that even if no new frontier models emerge over the next 3 years, significant progress can be made simply by deploying the technology that already exists.

In this situation, the industry's attention is shifting from competition over AI models themselves to the ability to secure infrastructure to scale them into real services. The pace at which bottlenecks such as memory chips and power are eased is expected to continue affecting the speed of data centre expansion and investment flows in related companies.