The rapid growth of AI is running into the much slower pace of building power infrastructure: a new report from Goldman Sachs warns that energy bottlenecks could impede the expansion of compute and data centers that underpin modern AI.
Why this matters
Against a backdrop of Wall Street concern about debt‑financed AI buildouts, Goldman Sachs identifies power availability as the largest risk to trillions of dollars in planned spending. Permitting and grid connection timelines are considerably slower than the pace at which companies are deploying advanced AI systems.
Key figures and findings
- Goldman Sachs expects global data center power demand to rise about 170% between 2025 and 2030, with more than 60% of that growth occurring in the United States.
- The bank estimates roughly $7.6 trillion in capital spending will be required across computing, data centers and power infrastructure between 2026 and 2031.
- In the U.S. there are roughly 2,600 gigawatts of proposed power generation and storage projects awaiting grid connections — more than twice the grid’s current installed capacity, though many projects may never be realized.
- The median time from a grid connection application to commercial operation is roughly five years.
Jason Tofsky, Goldman’s global head of digital infrastructure investment banking, summarized the issue succinctly: “Power is probably the forever bottleneck.”
Financial and operational risks
- Goldman argues that the so‑called AI race has increasingly become a contest over who can secure sufficient power. That dynamic could produce longer, less predictable buildouts, delayed returns and heightened risks for firms that invest heavily before ensuring power access.
- Much of the expansion is debt‑financed: Goldman projects $420 billion of hyperscaler debt issuance in 2027, more than 60% higher than its 2026 estimate.
- Debt is expected to finance roughly 35% of hyperscalers’ capital expenditures next year. Hyperscalers account for only about 40% of AI‑ecosystem credit fundraising, indicating that borrowing spans the broader industry.
Delays, local opposition and cancellations
- Since January, Goldman reports that 86 projects have been canceled or significantly delayed, and 76 of those disruptions are directly tied to local opposition. Together these trends point to extended implementation timelines.
- Slower deployment could, however, mitigate the risk of over‑investment by allowing actual demand to catch up with supply and spending.
Who benefits and how the sector shifts
- The AI buildout is turning utilities into major beneficiaries: utility capital plans are expected to reach about $1.3 trillion between 2026 and 2030.
- Some tech giants are exploring direct funding of power generation, transmission lines and electrical infrastructure to secure the energy their data centers require.
Bottom line
Goldman Sachs concludes that while AI systems attract users, sustained growth depends on having enough electricity to power them. Insufficient or delayed power capacity could limit how quickly and reliably AI deployments can scale.
