Powering AI: Takeaways from a Data Center Industry Conference
Summaries are public. PDF access requires an active membership; all members have the same access. Sign in
Bernstein published key takeaways from the DCD>Connect New York 2026 conference, highlighting intensifying energy bottlenecks, rising rack densities, and architectural shifts in AI infrastructure. On-site power generation emerged as a central theme. While only 13% of the 5,400 US data centers currently utilize on-site generation as a primary source, industry participants estimate that up to 40% of new facilities could operate off-grid. Delivery lead times for on-site generation equipment now extend into 2028. The 800V direct current ecosystem is rapidly maturing beyond the conceptual phase. Eaton showcased its end-to-end architecture featuring solid-state transformers, with 15kV deliveries beginning in Q4 2026 and a 35kV option currently under development. Meanwhile, Hubbell's PowerGain product eliminates hardwiring requirements, reducing deployment time by 30% and lowering overall capital and operating expenditures. Labor shortages represent another critical, yet underappreciated, bottleneck. Referred to as AI's sixth layer, skilled labor constraints could severely limit construction speed. Panelists warned that the US could face a shortage of up to 420,000 skilled technicians by 2030. With 22% of American electricians currently over the age of 55, replacing retiring workers will require training over 260,000 new electricians by 2034. Bernstein maintains an Outperform rating on Eaton with a target price of $428 and on Hubbell with a target price of $553, citing their competitive advantages in electrical equipment innovation. Conversely, Caterpillar, Cummins, and Quanta Services are rated Market-Perform, with target prices of $678, $600, and $515, respectively, reflecting more neutral exposure to these structural trends.