"Power is probably the forever bottleneck": Grid delays threaten the AI race.

Silicon Valley's rapid progress often clashes with the slow-moving U.S. electrical grid, which necessitates permits for development. Goldman Sachs warns that the ever-expanding AI industry is encountering bottlenecks in the power infrastructure buildout, potentially impeding the AI race. The warning comes from a new report exclusive to Axios.

Among the concerns highlighted is power as the chief risk to the vast investments in AI technologies. Goldman Sachs anticipates a staggering 170% increase in global data center power demand from 2025 to 2030, with over 60% of that surge originating in the United States. The bank projects a total of $7.6 trillion in capital spending across computing, data centers, and power infrastructure between 2026 and 2031.

A significant hurdle to this growth is the time lag between project approval and actual operational commencement, averaging around five years. Approximately 2,600 gigawatts of power generation and storage projects are currently pending grid connections, more than double the existing grid capacity, although many may not be realized.

Goldman argues that the AI race is ultimately about securing sufficient power, highlighting the potential for protracted construction timelines, delayed financial returns, and heightened risks for companies investing heavily before securing adequate power supply. The heavy reliance on debt financing exacerbates these risks, with Goldman forecasting $420 billion in global hyperscaler debt issuance in 2027, a more than 60% jump from 2026 projections.

Hyperscalers are projected to account for roughly 35% of capital expenditures in this sector next year, although they represent only 40% of AI ecosystem credit fundraising. Notable delays include 86 projects that have either been canceled or significantly postponed, with 76 of these attributed to local opposition. This multitude of delays underscores the extent of the challenges faced.

While a slower deployment pace might allow time for demand to catch up with spending, preventing potential over-investment, the boom is fundamentally altering the utilities sector, turning them into significant beneficiaries of AI investment. Capital plans for utilities are expected to reach $1.3 trillion between 2026 and 2030, as tech giants consider direct funding for the power generation, transmission lines, and electrical infrastructure essential for their data centers.

The bottom line remains that the availability of power is crucial for the success and sustainability of AI systems.