Master | Moments | Getty Images
Semiconductor stocks have rallied wildly over the past year as investors bet on the sector’s central role in building global AI infrastructure.
But volatility surrounding semiconductor stocks is rising again, sparking debate over whether this is a sign of broader concerns about demand for AI.
In interviews with CNBC this week, several AI executives poured cold water on the idea that demand is slowing, while acknowledging that companies are becoming more cautious about the cost of using AI.
“I think the demand for AI is almost limitless,” says Pat Gelsinger. intel Playground Global’s CEO and current general partner told CNBC on Wednesday, adding that energy availability is “the only real limiting factor.”
“Because how much economic value can be gained from increased intelligence? It’s almost infinite, across every conceivable industry,” Gelsinger added.
Data center reports supply constraints for chip players
Due to various factors, market volatility is increasing, especially in stocks related to chips and AI data centers. Meta’s announcement that it would sell excess AI computing power also contributed to the decline. While Meta’s stock soared on the news, it raised questions about whether this was a sign of a broader glut of computing power. Elon Musk’s xAI also lent out excess capacity this year.
And this week, Samsung, one of the world’s largest memory chip companies, predicted a big profit increase, but its stock price fell. After the stock rose more than 360% in the past 12 months, the market wondered how far the stock could go.
None of these moves seem to dampen the demand for computing and the infrastructure behind it.
“What we’re experiencing in terms of demand is extraordinary. There’s more demand than we can meet, and that’s been our experience for some time now,” said Chief Revenue Officer Mark Boroditsky. Neviushe told CNBC on Thursday. Nebius builds data centers using NvidiaGPU.
CEO Andrew Feldman cerebral systemI gave an example. Meta It is a “unique” case that xAI is selling excess capacity.
“As an industry, the demand for computing far exceeds the available capacity, and we’re running out of data centers. I think the industry as a whole is missing a lot of the inputs that computing requires,” Feldman told CNBC on Wednesday.
Cerebras, which went public earlier this year, is one of a number of semiconductor startups looking to become a major player in the data center market and challenge Nvidia.
Rebellions, another South Korean semiconductor startup backed by Samsung and SK Hynix, reported similar strong demand.
“The momentum in AI infrastructure remains strong,” Rebellions CEO Sungyun Park told CNBC on Wednesday.
“Personally, I believe this is not a signal that all hyperscalers are[overinvesting]in infrastructure,” Park added, referring to the meta and xAI news.
lumenThe company, which sells photonics and optical products for data center connectivity, said its products will be sold out for the next five years.
Lumentum CEO Michael Hurlston told CNBC on Wednesday: “We are looking to add as much capacity as we can to meet the demand that we currently see five years into the future.”
Lumentum’s stock has risen about 600% in the past 12 months as investors flock to companies that address key bottlenecks in building AI data centers.
“Rationalization” of corporate spending
Another big debate surrounding AI trade is how much companies are willing to pay for the technology.
There was a period of so-called “token maxing” in which companies encouraged employees to use AI as much as possible, regardless of the consequences. In many cases, tools from Frontier Labs such as OpenAI and Anthropic were used.
But especially these Frontier models, DeepSeek and alibaba.
Nebius’ Boroditsky said tokenmaxxing is only worthwhile if an organization can see a return on investment as a result.
“CFOs who put down the hammer and slow spending should actually be looking for value and value maximization,” Boroditsky said, adding that AI should be applied to create value that justifies spending.
“We’re currently seeing a shift towards further streamlining. We’ve seen it with every technology cycle, and that streamlining will definitely continue to be in demand,” Nevius’ Boroditsky said.
Although Frontier AI models are considered the most advanced, there are many open source models with similar or less advanced performance. Different models have different features available for specific tasks.
Cerebras’ Feldman said that in the future, specific models will be used in specific situations. For example, frontier models can be used for more advanced problems, but some workloads migrate to other workloads.
“I think we don’t need a giant bus to go to the grocery store,” Feldman said.
“Certain workloads will migrate to one type of computing, and easier workloads will migrate to other types. I think the same thing will happen as we learn to deploy AI and become more sophisticated.”
