A better way of thinking about the AI bubble

The conversation surrounding technology bubbles is often steeped in doom and gloom, but the reality may be less dire than it seems. In economic terms, a bubble represents a situation where investments exceed actual demand, leading to an oversupply. This situation is particularly poignant in the context of artificial intelligence, where the rapid advancements in AI software are at odds with the lengthy process of constructing and powering data centers. Building these data centers is a time-consuming process that can span several years. Consequently, the landscape may shift dramatically from the time investments are made to when these facilities become operational. The intricate and ever-evolving supply chain underpinning AI services complicates predictions about future demand and requirements. It’s not merely about how extensively AI will be utilized by 2028, but also about the manner in which it will be deployed and whether there will be significant advancements in energy efficiency, semiconductor technology, or power distribution in the interim. Recent developments highlight the scale of financial commitments being made in the AI sector. For instance, a data center campus in New Mexico, associated with Oracle, has secured an astonishing $18 billion in credit from a consortium of banks. Furthermore, Oracle has engaged in a $300 billion contract for cloud services with OpenAI, and in collaboration with Softbank, they are embarking on the ambitious “Stargate” project, which aims to establish $500 billion in AI infrastructure. Not to be outdone, Meta has announced a staggering $600 billion investment in infrastructure over the next three years. Despite these monumental promises, uncertainty looms over the pace at which demand for AI services will escalate. A recent McKinsey survey revealed a mixed bag of results regarding AI adoption among leading firms. While nearly all businesses surveyed reported some level of AI integration, only a handful are leveraging these technologies on a significant scale. AI has been beneficial for cost-cutting in specific instances, but its overall impact on business operations remains limited, with many companies adopting a “wait and see” approach. This could pose a challenge for data centers anticipating a surge in demand. Infrastructure challenges also threaten to hinder the progress of AI projects. Microsoft CEO Satya Nadella recently expressed concerns about a lack of available data center space rather than a shortage of chips, underscoring the critical nature of infrastructure in the AI landscape. Many data centers remain underutilized due to an inability to meet the power demands of the latest chip technologies. While companies like Nvidia and OpenAI are pushing forward rapidly, the pace of improvement in electrical grids and built environments lags behind, creating potential for significant bottlenecks in the future. For more insights into this evolving subject, tune into this week’s Equity podcast.