For months, Wall Street debated whether AI spending had become too much, too fast.

Now, after calls by some of the industry’s biggest names to slow frontier AI development, investors are asking whether AI spending could slow, too.

“Does a “pacing slowdown” necessarily imply a spending slowdown? We don’t think so,” Bernstein analysts wrote Monday, referencing Anthropic CEO Dario Amodei’s call to slow the pace of frontier AI development.

The AI trade isn’t over

Investors shouldn’t equate slower frontier AI development with weaker AI spending, Bernstein analysts wrote.

They added that Amodei is proposing to slow development from “extremely fast” to “only somewhat fast,” not halt it, while demand for AI compute is increasingly being driven by inference.

Bernstein said there is already insufficient AI computing capacity to meet demand, making it unlikely companies would materially scale back their AI spending plans.

The firm also argued that stronger AI safeguards could support long-term adoption by easing political and societal concerns about the technology.

Bernstein continues to favor Nvidia, Broadcom, and semiconductor equipment makers.

The case for enterprise software

If AI spending holds up, investors’ next question is where that money flows.

ISI Evercore said the next leg of the AI trade could favor the software that helps businesses put AI to work.

Evercore notes that most software companies are consumers of AI models rather than developers of them, meaning their success depends more on deploying models inside businesses than on how quickly frontier AI improves.

“In our view, the enterprise AI race is increasingly about turning model intelligence into secure, governed, repeatable workflows, not simply having access to the smartest model,” the analysts wrote.

If frontier AI development becomes “somewhat less frenetic,” the relative value of software that helps companies deploy, govern, and secure AI could increase, they wrote.

That could benefit enterprise software companies as well as cybersecurity firms, since AI agents will still need identity management, governance, monitoring, and security regardless of how quickly frontier models improve.

Don’t chase the first move

On Monday, the iShares Expanded Tech-Software Sector ETF gained 5% while the iShares Semiconductor ETF fell 6% as investors bet a slower pace of frontier AI development would favor software over AI infrastructure.

RBC Capital Markets analysts say the market’s initial reaction overlooked an important tradeoff.

While slower frontier AI development may give AI laggards more time to catch up, it could also delay the rollout of more capable AI products that many software companies are counting on to drive future growth.

“The market is pricing near-term safety while overlooking the opportunity cost of deferred AI adoption,” the analysts wrote.

If the slowdown proves temporary, companies whose products depend on frontier AI advances could have more upside than investors currently expect, they added.

Watch the next leg of the AI trade

The market reaction reflects changing expectations more than changing fundamentals, Charu Chanana, Saxo’s chief investment strategist, wrote on Monday.

“For now, this looks more like a sentiment and valuation shock than a collapse in AI demand,” she wrote.

Calls for more AI testing shouldn’t be mistaken for a pullback in AI investment, she added.

“More testing does not mean technology companies will suddenly stop building data centers or buying computing equipment,” she wrote.

However, if AI companies ultimately release fewer frontier models or run fewer massive training programs, demand for the most advanced processors, memory chips, and chip-packaging capacity could fall short of today’s expectations.

Memory chips could be particularly vulnerable because manufacturers are adding new supply to meet expectations of continued strong AI demand.

“If demand is delayed just as new production arrives, shortages could turn into excess supply and weaker prices,” Chanana wrote.