AI Executives See “Almost Unlimited” Demand Amid Stock Volatility

Semiconductor stocks have experienced a dramatic surge over the past year, fueled by investor confidence in the sector’s pivotal role in constructing the global artificial intelligence infrastructure. However, recent volatility in chip stocks has ignited a discussion about whether this trend signals a broader concern regarding AI demand.

In recent interviews, several AI executives have largely dismissed the notion of slowing demand, while acknowledging a growing emphasis on cost-efficiency in AI implementation by businesses.

“I view AI demand as bordering on limitless,” stated Pat Gelsinger, former CEO of Intel and now a general partner at Playground Global. He told CNBC, “The primary constraint, quite frankly, is energy availability.”

Gelsinger elaborated on the immense economic value generated by enhanced intelligence, asserting, “The economic upside from increased intelligence is virtually infinite across every conceivable industry.”

Several developments have contributed to the recent market fluctuations impacting chip and AI data center-related equities. Meta’s announcement that it plans to monetize its surplus AI computing capacity, for instance, played a role in the sell-off. While Meta’s stock saw an uptick on the news, it raised questions about the potential for broader compute overcapacity in the market. Similarly, Elon Musk’s xAI has also been observed to be leasing out its excess capacity this year.

Adding to this, Samsung, a global leader in memory chip manufacturing, recently projected a substantial increase in its second-quarter profits. However, its stock experienced a decline, with the market questioning its potential for further growth after an impressive rally of over 360% in its shares over the past twelve months.

Despite these market movements, the underlying demand for compute power and its supporting infrastructure appears undiminished.

“What we’re witnessing in terms of demand is truly extraordinary. We’re experiencing significantly more demand than we can currently fulfill, and this has been our reality for some time now,” commented Marc Boroditsky, Chief Revenue Officer at Nebius, a company building data centers utilizing Nvidia’s Graphics Processing Units (GPUs). He shared these insights with CNBC.

Andrew Feldman, CEO of Cerebras Systems, a company challenging Nvidia’s dominance in the data center market, characterized the instances of Meta and xAI leasing out their excess capacity as “unique.”

“For the industry as a whole, the demand for compute power significantly outstrips the available capacity, and we are experiencing a shortage of data centers. Furthermore, as an industry, we are facing shortages in many of the essential components required for compute,” Feldman told CNBC.

Cerebras, which recently went public, is among a wave of semiconductor startups aiming to make a significant impact in the data center market and compete with established players like Nvidia.

Rebellions, a South Korean chip startup backed by industry giants Samsung and SK Hynix, has also reported robust demand.

“The momentum for AI infrastructure remains immense,” stated Sungyun Park, CEO of Rebellions. He added, “Personally, I do not believe this indicates that hyperscalers are overinvesting in infrastructure,” referring to the recent news from Meta and xAI.

Lumentum, a provider of photonics and optical products essential for data center connectivity, reported that its products are fully booked for the next five years.

“We are actively working to expand our production capacity to meet the demand we anticipate over the next five years,” Michael Hurlston, CEO of Lumentum, informed CNBC.

Lumentum’s stock has surged approximately 600% over the past year, as investors flock to companies addressing critical bottlenecks in the development of AI data centers.

A significant point of discussion within the AI sector revolves around the willingness of enterprises to invest in the technology.

Previously, a trend referred to as ‘tokenmaxxing’ was observed, where companies encouraged employees to extensively utilize AI tools, irrespective of the outcome. This often involved proprietary models from leading AI labs like OpenAI and Anthropic.

However, businesses are now increasingly prioritizing the return on investment (ROI) from their AI deployments. This shift is particularly pronounced as the cost of advanced, proprietary models remains high when compared to open-source alternatives offered by companies such as DeepSeek and Alibaba.

Nebius’ Boroditsky emphasized that ‘tokenmaxxing’ is only economically viable if an organization is realizing a tangible ROI. He commented, “When a CFO imposes spending controls, the focus should be on achieving value, or ‘valuemaxxing.’ AI applications must demonstrate a clear return that justifies the investment.”

“We are observing a transition towards more rationalized spending. This pattern is consistent with previous technology cycles, and this rationalization will ultimately sustain demand,” Boroditsky added.

While frontier AI models represent the pinnacle of technological advancement, a wide array of open-source models now offer comparable performance, with some even approaching parity. These diverse models possess distinct capabilities, making them suitable for specific tasks.

Cerebras’ Feldman suggested that in the future, specialized AI models will be deployed for targeted applications. For instance, frontier models might be reserved for highly complex problems, while other, less demanding workloads could be allocated to different computational resources.

“It’s akin to using a bicycle for a short trip to the grocery store instead of a large bus,” Feldman illustrated. “As we gain more sophistication in AI deployment and understanding, we will see certain workloads migrate to optimized compute platforms, while simpler tasks will be handled by more accessible solutions.”