Demand for artificial intelligence infrastructure remains far stronger than available supply, according to technology executives, even as companies become more disciplined about how they spend on AI.
The comments come as investors debate whether recent volatility in semiconductor stocks signals that the AI boom is beginning to cool. Industry leaders interviewed by CNBC argued the opposite, saying enterprise customers are not pulling back from AI. Instead, they are becoming more strategic about where they deploy it and how they measure its return on investment.
Chip stocks have surged over the past year as Wall Street poured money into companies expected to benefit from the global AI infrastructure race. However, recent market swings have fueled concerns that spending on data centers and AI hardware could be peaking. Several developments added to those concerns.
Meta announced plans to sell excess AI computing capacity, while Elon Musk's xAI has also rented out unused compute resources. Meanwhile, Samsung projected a sharp jump in profits, yet its shares still fell as investors questioned how much more upside remained after a rally of more than 360% over the previous year.
Despite those headlines, executives across the AI ecosystem insist demand has not weakened. Former Intel CEO Pat Gelsinger, now a general partner at Playground Global, described demand for AI as virtually limitless. "I somewhat think of AI demand as almost unlimited," Gelsinger told CNBC, arguing that electricity, rather than customer demand, is becoming the industry's biggest constraint.
"The only real limiter is energy availability," he said, adding that the economic value generated by increasing intelligence across industries is "almost infinite." That optimism was echoed by executives building the infrastructure behind today's AI boom.
Marc Boroditsky, chief revenue officer of Nebius, said the company continues to see more requests for AI computing power than it can satisfy. Nebius develops AI cloud infrastructure powered by Nvidia graphics processing units, or GPUs.
"What we're experiencing in terms of demand is extraordinary," Boroditsky told CNBC. "There's much more demand than we're able to fulfill, and that's been our experience for some time now."
Andrew Feldman, chief executive of Cerebras Systems, said recent announcements by Meta and xAI should not be interpreted as evidence of excess industry capacity.
Instead, he described those situations as unique while emphasizing that the broader AI market continues to face shortages of data centers, computing infrastructure and key hardware components.
"For the industry as a whole, the demand for compute far outstrips available capacity," Feldman said. South Korean AI chip startup Rebellions reported seeing similar trends.
Chief executive Sungyun Park said AI infrastructure momentum remains "huge," reflecting continued investment in new computing capacity worldwide. The demand extends beyond chips themselves.
Over the past year, many companies embraced what some executives call "tokenmaxxing," encouraging employees to use large language models as much as possible regardless of cost. Those deployments often relied on premium AI models from companies such as OpenAI and Anthropic.
Now, finance executives are demanding measurable business results. Boroditsky said organizations are shifting toward what he calls "valuemaxxing," prioritizing AI applications that generate clear returns rather than simply maximizing usage. "The CFO bringing the hammer down and slowing spend should actually be looking for value or valuemaxxing," he said.
He added that every major technology cycle eventually moves from rapid experimentation to more disciplined deployment. "We're seeing a shift now to more rationalization," Boroditsky said. "We've seen it with every tech cycle." Executives also expect businesses to become more selective about which AI models they use.
Rather than relying exclusively on the largest and most expensive frontier models, companies are increasingly expected to combine premium systems with lower-cost open-source alternatives such as those developed by DeepSeek or Alibaba.
Less demanding tasks can be handled by smaller models, while complex reasoning workloads continue to rely on cutting-edge AI systems. Feldman compared the trend to choosing the right vehicle for a task. "I think it's probably the case that you don't need a giant bus to go to the grocery store," he said.
