Goldman Sachs Projects $700B in US Grid Investment Through 2030 — And It Still Won't Be Enough

Goldman Sachs Research, citing EEI and company data, projects more than $700 billion in US grid investment between now and 2030 — transmission and distribution CapEx combined, rising from roughly $90B per year in 2024 to over $110B per year by 2030. The numbers are real. The trajectory is clear. The investment is happening.

And it will not solve the AI data center deployment problem this decade.

This is not a contradiction. It is a sequencing problem. Understanding the gap between "grid investment is accelerating" and "AI data centers can deploy on schedule" is the most important analytical distinction in infrastructure right now — and the one most consistently missed by capital allocators evaluating the sector.

THE MATH THAT DOESN'T WORK

Grid investment and grid availability are not the same thing. When Goldman Sachs projects $100B+ per year in transmission and distribution CapEx, that capital is going into planning studies, permitting processes, environmental reviews, equipment procurement, and construction cycles that take 5 to 10 years to translate into energised capacity. The money committed in 2025 powers the grid of 2032, not 2026.

The developers who paused projects this year — Crusoe's 1.8 GW Wyoming campus, the 85% of projects delayed across the sector — were not paused because capital was unavailable. They were paused because the physical infrastructure to deliver power to the site does not exist yet, and no amount of investment commitment changes that timeline.

THE NEOCLOUD SIGNAL NOBODY IS WATCHING

The pressure on US grid infrastructure is not coming only from hyperscalers. A second wave is hitting the same constrained supply: neoclouds.

Neoclouds are the new hyperscalers for grid purposes. They require the same HV connections, the same transformer specifications, the same substation infrastructure. They have less negotiating leverage with grid operators and less internal capability to manufacture their own supply chain (unlike Crusoe, which built three transformer manufacturing facilities). They are more exposed to the grid bottleneck, not less.

For European market intelligence, this matters. Neoclouds evaluating European deployment are underserved by the current advisory landscape, which is focused almost entirely on hyperscaler-scale projects. A 50–200 MW neocloud deployment in France has the same grid navigation complexity as a 1 GW hyperscaler campus — but without the in-house team to manage it.

WHY $700B DOESN'T HELP YOU IN 2026

The Goldman Sachs projection is bullish on the right things: the US will eventually build the grid its AI ambitions require. The $700B commitment reflects genuine political will, regulatory momentum, and utility capex cycles that are now firmly oriented toward large-load accommodation.

But the timeline mismatch is structural, not cyclical. Three constraints compound each other:

Interconnection study queues. Before a single dollar of grid investment reaches a specific data center site, the developer must complete an interconnection study — a process that takes 2 to 4 years in ERCOT and PJM and cannot be accelerated by capital. The $700B of investment does not skip this queue.

Transformer procurement. Grid investment requires transformers. Transformer manufacturing capacity is the binding constraint on how fast the $700B can actually be deployed. ABB and Siemens are booked 48–60 months out. GE Vernova is at 60+ months. The White House invoked the Defense Production Act in April 2026 specifically because domestic manufacturing cannot meet demand. The $700B investment is partially queued behind the same transformer shortage it is trying to solve.

Permitting and environmental review. New transmission lines in the US require NEPA review, state permitting, and frequently litigation. The average new high-voltage transmission line takes 7 to 10 years from planning to energisation. The $700B buys the process, not the outcome, on a compressed timeline.

WHAT THIS MEANS FOR EUROPEAN DEPLOYMENT DECISIONS

The Goldman Sachs $700B figure is a signal that the US grid will eventually catch up with AI demand. It is not a signal that US grid capacity is available now. For developers and funds who need to deploy AI compute infrastructure in the 2026–2029 window, the US grid constraint is a present reality, not a future risk.

France is the mirror image of this dynamic. RTE's grid connection process is open, deterministic, and operating on 12–24 month timelines. The constraint is transformer procurement — a solvable problem with EU second-tier manufacturers at 20–32 month lead times — not a decade-long interconnection queue. RTE has identified 5 fast-track sites with 4,800 MW of near-ready capacity. Brownfield industrial sites with existing HTB connections compress timelines further to 18–24 months.

The developers who will have operational AI compute capacity in 2027–2028 are the ones making European grid decisions now, not waiting for the US market to absorb its own investment cycle. Goldman Sachs's $700B projection is the strongest possible argument for European deployment — because it quantifies exactly how long the US grid will remain constrained, and exactly how wide the window is for alternative markets.

The window is open. The $700B tells you when it closes.