I went to one of Wall Street's biggest tech conferences. The plumbing behind AI dominated the conversation.

This week at the JPMorgan tech and media conference, attendees gathered to discuss a topic critical to the current AI boom: the infrastructure that supports it. The focus on the necessary foundations of AI technology overshadowed discussions about prominent players like SpaceX, OpenAI, and Anthropic potentially going public this year. Held at the Westin Boston Seaport District, the conference served as a hub for networking among AI companies, chip manufacturers, telecom firms, fintech businesses, and the financial institutions backing them. While the atmosphere featured the usual conference elements—coffee cup graveyards, branded tote bags, and a sea of blazers—the emphasis was decidedly on infrastructure development. David de Boltz, a managing director in JPMorgan's leveraged finance capital markets, noted the enormous capital requests his team is encountering, labeling them as “astronomical.” The demand for investment in AI infrastructure projects is robust, with forecasts predicting data center development costs could reach into the trillions. De Boltz highlighted the necessity of every financial market—equity, high yield, and private credit—to meet the capital expenditure (capex) requirements of hyperscalers like Google and Meta. A standout moment came during CoreWeave's presentation, where Brannin McBee, the co-founder and chief development officer, pointed out the bottlenecks in building out AI infrastructure, including the challenge of hiring skilled labor and securing necessary equipment. CoreWeave has already established 43 AI data centers as of the end of last year. Marc Ganzi, CEO of DigitalBridge, also stressed the importance of the underlying fiber networks that connect AI systems. He criticized the industry for overlooking investment opportunities in these critical connections. This sentiment was echoed by Meta's recent initiative to train fiber technicians in collaboration with CBRE. As the urgency to build data centers intensifies, the timeline for construction has become a pressing concern. Typically, it takes two to three years to complete a facility, which can be delayed by local opposition, according to industry experts. Data centers are essential for companies like Google and Meta, which require substantial computing power for their AI projects. However, the intricacies of real estate agreements for these data centers can be complex and risky. A senior banker remarked that not all leases are created equal, and the potential for companies to transfer leases to less financially stable businesses adds another layer of risk. On a positive note, companies involved in the AI infrastructure buildout are experiencing growth. Caterpillar, for instance, has seen its stock surge over 45% this year. The company reported a 7% increase in construction industry sales, driven largely by demand for nonresidential projects, including data centers. With a record backlog of $63 billion, the company indicates that demand is significantly outpacing supply. Nevertheless, industry leaders at the conference cautioned that the pace of AI development is rapid and unpredictable. De Boltz emphasized the need for flexibility and agility in the face of constant change, as the AI landscape is evolving at an unprecedented rate.