Key industry figures dismissed market concerns that demand for artificial intelligence infrastructure is slowing. They said companies have become more sensitive to the cost of using AI, but underlying demand for computing power and data centres remains strong.

CNBC reported on July 12 that volatility in semiconductor shares and AI data centre-related stocks has increased, but assessments in the field say supply shortages persist.

A key question is whether the market can interpret the recent pullback in chip stocks as a sign that AI demand is slowing. Meta's statement that it would sell spare AI computing power to outside parties was cited as one factor behind rising selling pressure. Shares have risen, but the market has raised questions about whether AI computing infrastructure has already entered a state of oversupply. Elon Musk's xAI has also leased spare computing power this year.

The industry did not see this as weakening demand across the broader market. Pat Gelsinger (팻 겔싱어), a former Intel chief executive and now a general partner at Playground Global, said he views AI demand as effectively "almost limitless". He said energy supply is the "only practical constraint" and judged the economic value created by higher intelligence across industry to be very large.

Nebius, which is building data centres using Nvidia graphics processing units, said demand continues to exceed supply capacity. Marc Boroditsky (마크 보로디츠키), Nebius chief revenue officer, said, "The demand we're experiencing is unusual" and "far more than what we can meet." He added that this has not only emerged recently but has continued for some time.

Cerebras Systems also viewed the Meta and xAI cases as exceptions. Andrew Feldman (앤드루 펠드먼), Cerebras Systems CEO, said demand for computing power across the industry far exceeds available capacity and that there is also a shortage of data centres themselves. He said several inputs that support computing infrastructure are also in short supply across the industry. Cerebras Systems, which listed this year, is one of the semiconductor startups challenging Nvidia in the data centre market.

Rebellions, a South Korean semiconductor startup backed by Samsung and SK Hynix, shared a similar view. Rebellions CEO Park Seong-hyeon (박성현) said momentum in AI infrastructure remains strong and that he does not see the Meta and xAI cases as a signal that hyperscale cloud providers' infrastructure investment is excessive.

Lumentum, which supplies optical communications products for data centre connectivity, pointed to even stronger demand. Michael Hurlston (마이클 헐스턴), Lumentum CEO, said the company's products are effectively sold out for the next 5 years and that it is focusing on expanding production capacity to meet visible demand. That suggests demand is strong for optical communications components, which are cited as bottlenecks in expanding AI data centres.

Companies' approaches to adopting AI are changing. For a period, workplaces saw the spread of so-called "token maxing", which encouraged using as much AI as possible regardless of results. Leading models such as those from OpenAI and Anthropic were mainly used. More recently, instead of costlier leading models, a trend has strengthened toward using open-source models from DeepSeek or Alibaba and weighing cost against performance.

Boroditsky said chief financial officers curbing spending is not simply about cutting back but about demanding "maximising value". That means AI spending must actually create value to be justified. He said companies are now moving into a more rationalised stage and that this shift will instead keep demand going.

AI model selection is also moving toward greater segmentation. Feldman said the industry will shift away from using the same model for every task, and toward using leading models for difficult problems while moving relatively simple tasks to other models. "You don't need a giant bus to go to the grocery store," he said, predicting a trend in which computing resources and models will be matched to specific tasks.

Ultimately, market attention appears to be shifting from whether AI demand itself is shrinking to how budgets are allocated across computing resources and models. Investment enthusiasm in semiconductors and data centres continues, but corporate customers are beginning to judge more strictly not how much AI they use but where they use it to deliver cost-effective results.