AI managers say they want ‘almost unlimited’ between stock fluctuations

Chip stocks had a spectacular rally last year as investors bet on the semiconductor sector’s key role in building a global AI infrastructure.

But the renewed volatility surrounding chip stocks has sparked debate over whether this is a sign of broader concerns about the need for AI.

In interviews with CNBC this week, many AI executives poured cold water on the idea that demand is slowing, as they acknowledged that businesses are increasingly wary of the costs of implementing AI.

“In some ways I think the demand for AI is almost limitless,” Pat Gelsinger, ex Intel The CEO and general partner now at Playground Global, told CNBC on Wednesday, adding that capacity availability is “the only real limitation.”

“Because how much economic value do you get from increased intelligence? It’s almost infinite in every industry you can think of,” Gelsinger said.

Several factors have increased volatility in the markets surrounding chip and data center AI-related stocks. An announcement from Meta that it will sell its excess AI computing capacity contributed to sales. When Meta stock appeared in the news, we raised questions as to whether this was a sign that there was a broader computer boom out there. Elon Musk’s xAI has also hired its biggest capacity this year.

And this week, Samsung, one of the world’s largest memory chip companies, predicted a big increase in profits, but its stock fell. After a more than 360% rally in its shares over the past 12 months, the market questioned how much more it could go on.

None of these moves seem to reduce the need for computing and the infrastructure behind it.

“What we’re experiencing in terms of demand is amazing. There’s more demand than we can meet, and that’s our time now,” said Marc Boroditsky, chief financial officer. Nebiushe told CNBC on Thursday. Nebius builds data centers using NvidiaGPUs.

Andrew Feldman, CEO of Cerebral Systemssays the example of Meta and xAI selling its excess capacity is a “unique” case.

“In the industry as a whole, the computing demand far outstrips the available capacity, and we’re short on data centers. I think we’re short, as an industry, on a lot of hardware that needs to be put together,” Feldman told CNBC on Wednesday.

Cerebras, which went public earlier this year, is one of the semiconductor startups trying to become major players in the data center market and challenge Nvidia.

Rebellion, another chip startup from South Korea, backed by Samsung and SK Hynix, reported seeing similar demand.

“Infrastructure momentum for AI [is] it’s still big,” Sungyun Park, CEO of Rebellions, told CNBC on Wednesday.

“I personally believe that it is not a signal that … all hyperscalers [are overinvesting] infrastructure,” Park added, referring to Meta and xAI issues.

Lumentumwhich sells photonics and optical products for data center connectivity, said its products will be sold for the next five years.

“We’re trying to build our capacity as much as we can to meet the demand we see five years from now,” Michael Hurlston, CEO of Lumentum, told CNBC on Wednesday.

Lumentum’s stock has risen nearly 600% in the past 12 months as investors flock to companies facing challenges in building AI data centers.

Another big debate surrounding the commercialization of AI is how much businesses are willing to pay for the technology.

There has been an era called ‘tokenmaxxing’ in business where companies will encourage employees to use as much AI as possible regardless of the outcome. Commonly used tools are those from frontier labs such as OpenAI and Anthropic.

But companies are now more focused on the return on investment from AI, especially since those types of parameters are always more expensive compared to open offerings from companies like DeepSeek or. Alibaba.

Nebius’ Boroditsky said that tokenmaxxing is only justified if the organization sees a return on investment as a result.

“A CFO who is lowering the hammer and reducing spending should actually be looking at value or valuemaxxing,” Boroditsky said, adding that AI should be used to create value that enables spending.

“We’re seeing a shift now to more intelligence. We’ve seen it throughout the technology cycle, and that improvement will continue to be sought,” said Nebius’ Boroditsky.

While frontier AI models are considered the most advanced, there are a number of open source models that are close to working and others are not as advanced. Different models have different capabilities, which can be used for specific tasks.

Cerebras’ Feldman said that in the future, certain models will be used in certain situations. For example, boundary models can be used for more advanced problems, while some tasks will carry over to others.

“I think you probably don’t need a big bus to go to the grocery store,” Feldman said.

“Some workloads are moving to one type of computing and light work to others, and I think as we learn and become more sophisticated in our deployment of AI, the same thing will happen.”