Goldman raises global data center capacity forecast to 273 GW by 2030

Goldman Sachs raised its forecast for global data center capacity by 2030 to 273 GW from 217 GW, implying 170 GW of additions from 2025 and an estimated $8.5 trillion in cumulative five-year capex. The U.S. is forecast to reach 152 GW and account for roughly 60%–70% of global additions; selected U.S. technology and cloud companies’ annual capex could rise to about $1.74 trillion by 2030 from $457 billion in 2025. Goldman also flagged power constraints, permitting delays, community opposition and higher financing costs as risks to delivery.

Analysis

The investable signal is not the headline capacity number; it is the widening gap between announced demand and power-ready, financeable projects. Capacity forecasts can rise faster than energized capacity, so value should accrue first to constrained inputs—grid connection, transformers, switchgear, backup generation and power-aware construction—not uniformly to data-center landlords or developers. Screen Eaton, Vertiv, Quanta Services and GE Vernova for backlog quality and delivery capacity rather than treating the forecast as equivalent to booked revenue.

The second-order risk sits with cloud buyers: if infrastructure spending compounds faster than AI workload monetization, depreciation and financing costs can pressure free cash flow and returns on invested capital, even while equipment suppliers benefit. High project leverage also makes developers vulnerable to higher rates, delayed energization and refinancing conditions. The cited capex outlook is a forecast, not committed spending; independently verify customer budgets, project financing, power contracts and equipment orders.

Near term, constrained-power exposure may attract a relative premium. Over 1–3 months, permitting, interconnection and financing disclosures should separate executable projects from pipeline. Over 6–18 months, watch utilization and AI revenue conversion: weak monetization could defer later build phases. The contrarian point is that the forecast may be directionally right on demand but overstate timely capacity additions—and thus understate scarcity value for permitted power while overstating broad-based data-center revenue. Falsifiers include sustained project cancellations or capex cuts, easing equipment backlogs, and evidence that AI workloads are generating returns sufficient to support planned spending.

Key Decisions for Investors

• Prefer a selective long basket of grid and data-center electrical-equipment suppliers (for example, Eaton, Vertiv, Quanta Services and GE Vernova) over a broad data-center theme. Enter in tranches; confirm orders/backlog and delivery capacity before adding. Thesis weakens if lead times and backlogs contract without corresponding revenue conversion.
• Consider a relative-value position: long power-ready infrastructure suppliers versus highly levered data-center developers with uncontracted power or unresolved permits. Keep sizing modest until project-level leverage, power commitments and cancellation terms are verified; rising financing costs or delayed interconnection are key catalysts for underperformance.
• Treat hyperscaler capex as a watch item, not an automatic long. Track quarterly capex guidance alongside cloud growth, AI-related revenue disclosure and free-cash-flow conversion; reduce exposure if spending rises while monetization or returns on invested capital deteriorate.
• Over the next 1–3 months, prioritize filings and disclosures on utility interconnection, permits, financing and equipment orders. If projects increasingly secure power and funding, broaden exposure; if announced capacity remains pipeline-only, avoid paying for headline growth.