U.S. grid constraints emerge as key risk to AI infrastructure spending

Traders Union
All News
Business & Wealth
Companies
U.S. grid risk for AI infrastructure

Highlights

Goldman Sachs projects global data center power demand to surge 170% from 2025 to 2030, with the U.S. driving over 60% of the increase.
About $7.6 trillion in capital spending is anticipated across computing, data centers, and power infrastructure between 2026 and 2031, with grid interconnection timelines averaging five years.
Goldman expects global hyperscaler debt issuance of $420 billion in 2027, utilities' AI-related capital plans of $1.3 trillion through 2030, and 86 projects canceled or delayed since January.

Power demand surge tests infrastructure timelines

As first reported by Axios, Goldman Sachs says global data center power demand is expected to jump 170% between 2025 and 2030, with more than 60% of that increase coming from the U.S. In the report, Jason Tofsky, Goldman’s global head of digital infrastructure investment banking, says power is likely to remain the sector’s lasting bottleneck.

The bank estimates about $7.6 trillion in capital spending between 2026 and 2031 across computing, data centers and power infrastructure. Goldman argues that the race to scale AI is no longer defined only by advances in models, but increasingly by access to enough electricity to support them.

The report says the U.S. has roughly 2,600 gigawatts of proposed power generation and storage projects waiting for grid connections, more than twice existing installed capacity, although many of those projects may not move forward. It also puts the median period from a grid interconnection application to commercial operation at about five years.

Debt exposure and utility upside grow

The financing structure behind the expansion is adding to the risk if construction schedules slip. Goldman projects global hyperscaler debt issuance of $420 billion in 2027, more than 60% above its 2026 estimate, and says debt is expected to fund about 35% of hyperscaler capital expenditures next year.

Goldman also says hyperscalers account for only 40% of AI ecosystem credit fundraising, indicating that borrowing is spread widely across the sector. Since January, 86 projects have been canceled or significantly delayed, including 76 tied directly to local opposition, a pattern the bank says points to prolonged development hurdles.

At the same time, slower deployment could reduce the risk of over-investment by giving real demand more time to catch up with spending. Utilities are emerging as major beneficiaries of the AI buildout, with capital plans expected to reach $1.3 trillion between 2026 and 2030, while large technology companies are considering direct funding arrangements for generation, transmission lines and other electrical infrastructure needed for their data centers.

In our earlier article on the U.S. AI data center boom, we highlighted how rising financing costs, construction delays, and power constraints are undermining developers’ ability to deliver projects on schedule. We also noted that only a minority of planned builds were viewed as truly credible, with local opposition and grid limitations increasingly forcing cancellations, redesigns, or slower rollouts.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.