A diagnosis found that while general corporations have begun to manage artificial intelligence (AI) model usage expense more strictly than before, demand for AI infrastructure itself still exceeds supply.
U.S. business network CNBC reported on Dec. 12 (local time) that, based on interviews with former Intel Chief Executive Officer (CEO) Pat Gelsinger and executives at AI infrastructure corporations, corporations have only begun to scrutinize the efficiency of AI investment, while AI demand remains solid.
Gelsinger, now CEO of venture capital firm Playground Global, said, "AI demand is practically close to unlimited," and added, "The only constraint is power supply." He added, "The economic value AI will create is close to infinite across nearly every industry."
Mark Boroditsky, chief revenue officer (CRO) at Nevious, a neocloud corporation that builds Nvidia graphics processing unit (GPU)-based data centers and rents them to AI startups, also said, "The demand we are experiencing now far exceeds what we can handle," adding, "This situation has already continued for quite some time."
Recently, as Meta said it could rent out surplus AI computing resources to external parties and xAI also engaged in lending excess capacity, the market raised concerns that AI infrastructure might be oversupplied. The share prices of AI Semiconductor-related stocks such as Samsung Electronics and SK hynix were also cited as having wobbled for this reason.
However, the industry views this as a temporary phenomenon. Andrew Feldman, CEO of AI chip corporation Cerebras Systems, said, "The cases of Meta and xAI are exceptional," adding, "Across the industry, demand for compute resources far exceeds supply capacity, and data centers and compute infrastructure are still in short supply."
Rebellions, an AI Semiconductor startup backed by investments from Samsung Electronics and SK hynix, offered a similar assessment. Park Seong-yoon, Rebellions' CEO, said, "The momentum to build AI infrastructure remains strong," adding, "The renting of computing resources by Meta and xAI does not mean hyperscalers have overinvested."
Optical communications equipment maker Lumentum also said demand for AI data centers continues to expand. CEO Michael Hulsen said, "Most of what we will produce over the next five years has already been sold," adding, "We are expanding capacity as much as possible to meet demand five years from now."
However, assessments say AI investment approaches are changing. Rather than blindly increasing AI model usage as in the past, organizations are moving to a stage where they examine return on investment (ROI).
CRO Boroditsky explained, "Corporations' chief financial officers (CFOs) have started to scrutinize AI budgets more strictly," but added, "This does not mean cutting AI investment; it is a shift to a 'value maximizing' phase that maximizes investment efficiency."
CEO Feldman predicted that, going forward, the practice of selecting AI models based on the nature of the task will spread. He said, "State-of-the-art AI models will be used for complex tasks, and lightweight models will handle simple work," adding, "It's the same principle as not using a large vehicle every time you go around the neighborhood."