AI executives say demand remains strong as customers scrutinize costs

AI infrastructure companies are telling investors the buildout still has plenty of demand behind it, despite recent swings in chip and data center stocks. For everyday investors watching Nvidia-linked trades, the key tension is clear: companies still want more AI capacity, but customers are starting to ask harder questions about what they get for the money.

Several executives told CNBC this week that they are not seeing broad signs of excess capacity in artificial intelligence infrastructure. Compute, the processing power used to train and run AI models, remains scarce across parts of the market, they said.

Pat Gelsinger, the former Intel CEO who is now a general partner at Playground Global, told CNBC on Wednesday that he views AI demand as “almost unlimited.” He said power availability is “the only real limiter,” because he believes added intelligence can create economic value across industries.

Volatility has put the AI buildout under the microscope

Chip stocks have rallied sharply over the past year as investors bet that semiconductors will sit at the center of AI infrastructure spending. Recent volatility, though, has revived questions about whether the market is pricing in too much demand.

CNBC cited several triggers for those questions. Meta said it would sell excess AI computing capacity, and xAI rented out surplus capacity earlier this year. Meta’s shares rose on its announcement, but the move still raised concerns about whether some large AI buyers had built more capacity than they currently need.

Samsung also added to the debate. The memory chipmaker forecast a major increase in profit, according to CNBC, but its stock fell after a gain of more than 360% over the prior 12 months, as investors questioned how much upside remained.

Executives tied to AI infrastructure said those examples do not show a broad demand slowdown. Marc Boroditsky, chief revenue officer at Nebius, told CNBC on Thursday that demand is running ahead of what the company can supply. Nebius is building data centers that use Nvidia graphics processing units, or GPUs, the chips commonly used for AI workloads.

Andrew Feldman, CEO of Cerebras Systems, told CNBC that Meta and xAI renting or selling excess capacity looked like unusual cases. He said the broader industry still has more demand for compute than available supply and remains short of data centers and other inputs.

Sungyun Park, CEO of South Korean chip startup Rebellions, also told CNBC that AI infrastructure momentum remains strong. Park said he does not view the Meta and xAI moves as evidence that cloud giants, often called hyperscalers, are overinvesting across the board.

Lumentum, which sells photonics and optical products used for data center connectivity, pointed to another constraint. CEO Michael Hurlston told CNBC that the company’s products are sold out for the next five years and that Lumentum is trying to add capacity to meet demand. CNBC reported that Lumentum shares are up around 600% over the past 12 months.

Customers are focusing more on AI returns

The spending debate is shifting from raw usage to return on investment, meaning whether the money spent produces enough value to justify the cost. CNBC reported that some companies had gone through a period of “tokenmaxxing,” where employees were encouraged to use as much AI as possible, often through advanced models from companies such as OpenAI and Anthropic.

Boroditsky told CNBC that heavy AI use only makes sense when it produces a return. He said finance chiefs should be looking for value, using the term “valuemaxxing,” and applying AI in ways that justify the spending.

That shift also affects which AI models companies choose. CNBC noted that advanced frontier models can be more expensive than open-source options from companies such as DeepSeek or Alibaba, while some open models are close in performance for certain uses.

Feldman told CNBC that different AI jobs will likely move to different kinds of models and compute. More advanced problems may still call for frontier models, while easier tasks may run on cheaper alternatives. For investors, that means AI demand may remain strong while spending becomes more selective.

This story draws on original reporting from CNBC.