The heads of the world’s most powerful artificial intelligence companies want to slow down.

Data center developers don’t seem worried.

Following a string of security incidents involving frontier AI models and a wave of viral warnings about their dangers, the leaders of several major AI labs have spent the past few days pledging restraint.

Anthropic CEO Dario Amodei kicked it off with an essay published Saturday, writing that “we must slow the pace at which we improve the capabilities of AI models.” OpenAI’s Sam Altman and SpaceXAI’s Elon Musk both endorsed the call within hours, while Google DeepMind’s Demis Hassabis backed the essay’s direction as “correct for meeting this critical moment.”

Amodei said Anthropic would add new safeguards and let independent evaluators oversee its systems. Altman went further, telling Fortune the same weekend that OpenAI shouldn’t go public in 2026, calling it “an ill-advised moment” given the safety landscape.

Investors took the warnings seriously. AI stocks sank across global markets this week — the Nasdaq slid to a six-week low, with chipmakers dropping the hardest, and SoftBank plunged more than 13% in Tokyo — as fears spread that a slowdown could choke off the demand driving the industry’s spending boom.

But there is little panic emanating from within the data center sector itself.

Industry insiders who spoke with Bisnow expressed doubt that a true moratorium on model development would happen at all, despite the rhetoric from AI executives. But even if it does, they say the worst-case scenario for data centers falls well short of a crisis.

“Based on the discourse over this weekend, it appears like there's a consensus among tech CEOs that there needs to be a slowdown in the training of frontier models, but actions speak louder than words, and the companies that are being led by these tech CEOs are still planning their data center build-outs for the next five to 10 years,” said John Andril, a marketing intelligence manager for Avison Young specializing in critical infrastructure. “I don't see any impact on demand today.”

Despite pledges from the heads of frontier labs, experts who spoke with Bisnow were skeptical that any significant shutdown of AI model development is likely.

The economics and geopolitics of the AI arms race make a coordinated slowdown difficult to achieve, Andril said. He likened the situation to a game theory problem: A pause would require every frontier AI lab to stop developing new models simultaneously, with each firm having full faith that its competitors were abiding by their commitment. Given the intense competition between these firms, the chances of this kind of coordinated slowdown are remote.

Andril also noted that talk of a slowdown is only happening among leaders of U.S. AI firms. For a pause on model development to be effective, it would require buy-in from AI firms globally, particularly in China.

This doesn’t seem likely, Andril said. With Washington and Beijing locked in an AI arms race, the pressure to outpace rivals provides a strong incentive for both countries’ leading firms to continue pushing the advancement of frontier models, despite the risks.

“Chinese tech companies are not ignorant to the dangers posed by AI, but as of right now, we don't see tech CEOs over there making similar statements about a potential voluntary pullback, and there's no reason to believe that Chinese companies or the Chinese government would enter or adhere to such an AI disarmament,” Andril said.

Anthony Wanger, an industry veteran who leads data center investment firm Regnaw Capital and advises several companies, said he does believe that frontier labs will have to adopt a more cautious approach to model development as the number of security incidents mounts. But he cautions against overstating what such a slowdown would actually entail and the degree to which demand for AI computing would actually be impacted.

The talk of an AI pause, he said, is largely confined to frontier AI labs like OpenAI and SpaceXAI, not hyperscalers like Meta, Microsoft, Amazon and Google that account for the lion’s share of data center demand.

Rather than a fundamental shift in the trajectory of the AI industry, Wanger said the promises of restraint reflect the natural maturation of the major AI startups as they evolve into large companies that will soon be public. What is being framed as a slowdown, he said, is frontier labs beginning to adopt the same governance and oversight practices that have long been standard at the largest tech firms, and at major corporations in general. That’s hardly slamming the brakes on AI development, he said.

“They’re talking about pacing model development. They’re not talking about not doing it,” Wanger said. “Anthropic and OpenAI are growing up. It’s really hard to be a large public company in America.”

But even if the frontier labs were to completely halt the development of new AI models, the broad belief within the data center industry is that the repercussions for the sector would be limited. A pause would primarily affect firms building facilities specifically for AI training, which accounts for just a segment of data center demand.

The broader data center market would continue to be supported by other sources of demand: the traditional cloud services and enterprise workloads that remain a cornerstone of the industry, and AI inference workloads. Greater AI adoption is increasing the need for inference computing, which requires smaller data centers closer to end users.

The limited share of demand represented by AI training was emphasized this week by Digital Realty CEO Andrew Power.

Data Center REITs Digital Realty and Equinix, the world’s two largest data center companies, both saw their share prices drop Monday after the calls for an AI slowdown. But Power said his company would be largely unaffected if such a slowdown were to take place, as the firm has relatively little exposure to frontier model training.

“There’s tremendous digital transformation happening that is not connected to AI,” Power told CNBC. “There is tremendous cloud computing growth. Frankly, from my business lens, my seat, I think those demand trends, which are massive drivers of our business, have been stifled in these days of AI.”

Still, the share of data center demand driven by AI training isn't negligible. Avison Young’s Andril estimated that between 10% and 20% of current and projected data center demand is tied to training frontier models.

For any other sector of CRE, losing such a large chunk of demand would be “catastrophic,” Andril said, yet for the data center industry, it’s possible that even the evaporation of this much of the demand pipeline “would not really even be felt.”

The reason is what Wanger calls “a profound imbalance between supply and demand.” Despite the unprecedented AI building boom, demand for data center capacity continues to far outpace the industry’s ability to deliver it.

Most data centers in development today are being built to meet demand that already exists, with the vast majority of new capacity committed months or even years before the facilities come online. At the same time, tech giants like Microsoft and Amazon have repeatedly told investors that their record infrastructure spending is required to catch up to customer demand that they have already secured.

This demand backlog, along with the yearslong timeline to plan and build a data center, means that much of the infrastructure needed to support AI over the next five to 10 years is already being financed or under construction, both Wanger and Andril said. With so much future demand effectively locked in, a potential pause in model development being discussed by tech CEOs on social media will likely have little immediate impact on the trajectory of the data center sector.

“There are so many supply constraints in place and such a large backlog, and that backlog would probably exist even in the absence of frontier training demand,” Andril said. “It honestly might even be a relief to be able to develop data centers at just a slightly more relaxed pace than has been happening over the last five years.”