Over the past year, the semiconductor sector has experienced a remarkable upswing, driven by investor optimism surrounding its pivotal role in developing global AI infrastructure. However, this recent surge in volatility among chip stocks has sparked discussions about the potential slowing demand for AI. In conversations with CNBC this week, several AI leaders dismissed concerns about diminishing demand, while acknowledging that businesses are becoming more judicious in their AI expenditures. Pat Gelsinger, former CEO of Intel and now a general partner at Playground Global, stated, "I view AI demand as almost endless," noting that energy availability is the primary constraint. He further emphasized that the economic value derived from enhanced intelligence is "nearly infinite across all industries."
Several factors have contributed to the fluctuations in markets related to chip and AI data center stocks. One significant event was Meta's announcement of plans to sell its surplus AI computing capacity, which, while boosting its stock, raised concerns about potential overcapacity in the market. Similarly, Elon Musk's xAI is also leasing out its excess capacity this year. Moreover, Samsung, a leading memory chip manufacturer, forecasted a substantial profit increase, yet its stock dipped, leading to speculation about the sustainability of its nearly 360% stock rally over the past year. Despite these developments, demand for computational resources and the supporting infrastructure remains robust. Marc Boroditsky, chief revenue officer at Nebius, shared with CNBC on Thursday that the company is witnessing exceptional demand that surpasses their ability to meet it.
Andrew Feldman, CEO of Cerebras Systems, characterized the examples of Meta and xAI as "unique" scenarios. He pointed out that, for the industry overall, the demand for computational power exceeds the current capacity, indicating that there is a shortage of data centers. Cerebras, which went public recently, is among a wave of semiconductor startups aiming to capture significant market share in data centers and challenge giants like Nvidia. Sungyun Park, CEO of Rebellions—a chip startup backed by Samsung and SK Hynix—voiced similar sentiments, asserting that the demand for AI infrastructure remains substantial and that recent announcements from Meta and xAI do not signal overinvestment among hyperscalers.
Furthermore, Lumentum, known for its photonics and optical connectivity products for data centers, reported that its offerings are sold out for the next five years. CEO Michael Hurlston remarked, "We're striving to enhance our capacity to meet a demand that we project will last five years." Lumentum's stock has surged approximately 600% over the past year as investors gravitate towards companies addressing critical issues in the AI data center expansion.
Another pressing issue within the AI sphere is how much enterprises are willing to invest in technology. Some organizations have indulged in 'tokenmaxxing', encouraging extensive use of AI tools regardless of the outcome. However, there is a shift occurring as companies reevaluate the return on investment from AI technologies, especially as cutting-edge models from firms like OpenAI and Anthropic are often costly compared to open-source alternatives from companies such as DeepSeek or Alibaba. Boroditsky from Nebius remarked that 'tokenmaxxing' should only be pursued when it results in tangible returns. He added that CFOs need to prioritize value maximization as they curtail spending, suggesting that companies are transitioning toward more rational approaches to technology investment. This trend, he believes, will continue to drive demand in the AI sector.
