Entogo says power equipment, not compute, is now the AI data center bottleneck

By AI, Created 4:46 AM UTC, June 01, 2026, /AGP/ – A new Entogo analysis says North American AI data center timelines are being set by transformer and substation lead times, not chips. The company points to DOE data, surging electricity demand and limited power-equipment capacity as the main reasons projects are waiting years to energize.

Why it matters: - AI data center schedules in North America are increasingly being determined by power equipment availability rather than compute procurement. - The shift affects when projects can energize, which can delay revenue, capacity additions and utility interconnection plans. - Lead times for large transformers now reach roughly one and a half to four years in the merchant market, with substation packages taking longer.

What happened: - Entogo published a new analysis this week on why sourcing power equipment now sets the timeline for AI data centers. - The analysis cites the U.S. Department of Energy’s July 2024 Large Power Transformer Resilience Report, which puts current lead times for large power transformers at 80 to 210 weeks. - Entogo says the constraint is now power equipment, not compute. - The company is based in Toronto and serves North American and global markets.

The details: - Grain-oriented electrical steel, or GOES, is a core input for every power transformer and represents about a quarter of large-transformer cost in the DOE’s accounting. - Only a handful of mills worldwide produce GOES. - Lawrence Berkeley National Laboratory’s December 2024 study for the DOE estimates U.S. data centers used about 4.4% of national electricity in 2023. - The same study projects data centers will reach 6.7% to 12% of U.S. electricity use by 2028. - EPRI projects a similar demand trajectory. - An aging installed transformer fleet is also driving replacement demand that competes with new projects for manufacturing capacity. - Rack densities have moved from 10 to 15 kW, which defined enterprise IT for two decades, to about 120 kW for current NVIDIA Blackwell racks. - Open Compute Project reference designs reach 1 MW. - Those loads are harmonic-rich and run near full load around the clock. - Transformers feeding those loads must be specified with a K-factor under ANSI/IEEE C57.110. - Those transformers also need to be engineered for partial-load efficiency rather than peak rating. - In a 100-megawatt campus, the 33 to 35 kV campus collector layer carries most of the equipment specification and most of the lead-time risk. - Entogo’s companion analysis covers the voltage hierarchy from utility tie to rack, switchgear fault-level design for behind-the-meter generation and how battery storage complements, rather than replaces, the UPS layer.

Between the lines: - The analysis draws a sharp distinction between the merchant market and a vertically integrated manufacturer. - The merchant market is the queue buyers enter when ordering from a broker or from a factory already running at capacity. - Entogo argues that a manufacturer running its own factory with a vertically integrated supply chain can set delivery by its own production capacity instead of that backlog. - That framing turns procurement strategy into a competitive advantage for AI infrastructure developers.

What’s next: - Entogo says its own equipment is manufactured in a source factory, including transformers, prefabricated substations and medium- and low-voltage switchgear. - The company says it ships European-standard IEC/CE catalogue equipment in an average of 12 weeks. - Entogo says it guarantees delivery within 36 weeks even when a product needs new UL or other North American certification. - The company’s stated delivery window compares with an industry baseline of one to four years. - Entogo publishes its analyses at entogo.ca/insights.

The bottom line: - For AI data centers, the critical path is shifting from servers to substations.