Goldman Sachs raises 2030 global data center capacity to 217GW, but believes supply-demand will remain tight

Overview

This report updates Goldman Sachs' views on global data center capacity, power demand, and related stocks. Based on the 451 Research database, Goldman Sachs raises its forecast for 2030 global data center supply to 217GW, up 116GW from 101GW in 2025; US IT Power supply is expected to reach 125GW by the end of 2030. The report believes demand from AI, cloud computing, neocloud, and GPU-as-a-service is still expanding, and although new capacity is substantial, lower vacancy, rising lease prices, and power constraints mean supply-demand tightness will persist.

Core views

The core views include: first, global data center capacity expectations have been revised up significantly, but this is still insufficient to eliminate scarcity in compute and power resources; second, hyperscalers' average expected annual capital expenditures of about $1 trillion are enough to cover the capex required for capacity expansion, and this still does not include SpaceX AI capex, private AI lab capex, and data center platform capex; third, the neocloud business model is more financeable due to long-term leases, GPU service transactions, and Nvidia revenue guarantee/revenue-sharing models; fourth, US power demand growth has been revised up to a 3.5% CAGR through 2030, with PJM, MISO, ERCOT, and the Southeast becoming key regions; fifth, regulated utilities, independent power producers, and data center REITs all have clear paths to benefit.

Analysis framework

The report uses a combination of top-down and bottom-up approaches: it updates global and US capacity supply using 451 Research project tracking data; combines the TMT team's AI server shipment forecasts, 650 Group equipment capex estimates, CBRE vacancy rates, and channel checks to assess supply-demand tightness; in the US power section, it estimates data center demand, utilization, occupancy, PUE, and BTM and FTM power supply paths by regional power market; and for stock views, it incorporates regional exposure, capex opportunities, valuation multiples, and company-specific risks.

Methodology notes

• Supply forecast451 Research data center project tracking

  Global and US data center capacity forecast

  The report uses 451 Research data center project tracking data as of C1Q26 to raise its 2030 global capacity forecast to 217GW and estimate US IT Power supply at 125GW by the end of 2030.

• Demand constraint framework7 Ps

  Drivers and constraints of data center power demand growth

  GS SUSTAIN proposes seven driving or constraining factors, including AI adoption, chip/server/model productivity, electricity prices, policy, component availability, labor availability, and the physical environment.

• Regional power modelUS regional bottom-up demand estimation

  Estimate data center demand by markets such as PJM, MISO, and ERCOT

  Based on 121GW of IT Power, about 1.3 PUE, a 70% weighted average utilization rate, and a 97% occupancy rate, the report estimates US data center demand at 108GW/947TWh in 2030.

• Equity valuationP/E, EV/EBITDA, FCF yield, P/AFFO and SOTP

  Valuation methods for target prices of covered stocks

  Utilities use P/E or SOTP, independent power producers use a blended EV/EBITDA and FCF yield valuation, and DLR uses NTM+1Y P/AFFO.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Digital Realty Trust Inc. (DLR)

  A global data center REIT that directly benefits from tight data center supply-demand conditions, lease repricing, and AI/cloud demand.

  Strengths

  Owns more than 300 facilities across over 50 metro areas and serves nearly 6,000 customers; backlog and joint venture arrangements with institutional partners provide multi-year revenue and earnings visibility.

  Weaknesses

  The business is capital-intensive and still needs to manage capital requirements and customer concentration risk.

  Comparison

  Compared with pure power names, DLR has more direct exposure to data center leasing and REIT valuation; the report rates DLR Buy, while EQIX is shown as Neutral in the heading.

  Risks

  Key risks include cost of capital, customer concentration, weaker-than-expected renewal pricing realization, and slower data center construction or demand.

• Equinix Inc. (EQIX)

  A data center and interconnection services company exposed to the same data center supply-demand theme.

  Strengths

  The heading shows it is a covered name, and industry-wide tight supply-demand conditions are favorable for data center operators overall.

  Weaknesses

  The provided excerpt does not disclose detailed company-level investment rationale.

  Comparison

  The report heading shows EQIX as Neutral, while DLR is Buy.

  Risks

  The excerpt does not provide company-specific risks; general risks come from slower demand, cost of capital, construction, and competition.

• IREN Ltd (IREN)

  A neocloud/GPU-as-a-service-related name involved in GPU service transactions with Nvidia, Perplexity, Figure, and others.

  Strengths

  GPU service transactions show that large-scale compute capacity is scarce, while neocloud financing models and near-term capacity demand are improving.

  Weaknesses

  The heading shows IREN as Neutral, indicating that company-level risks or valuation may limit the rating.

  Comparison

  Compared with traditional data center REITs, IREN is more oriented toward compute services and GPU capacity business models.

  Risks

  Short-term leases, financing costs, GPU prices, customer demand, and declining compute pricing are potential risks.

• FirstEnergy Corp. (FE)

  A PJM-region regulated utility benefiting from concentrated data center demand and generation/transmission opportunities in West Virginia.

  Strengths

  Buy rating with a 12-month target price of $54; the report believes it benefits from PJM exposure.

  Weaknesses

  Affected by regulated returns, interest rates, and pension expenses.

  Comparison

  Among PJM beneficiaries, FE differs from IPPs such as TLN by being more of a regulated utility path.

  Risks

  Adverse outcome in the Ohio rate case, higher-than-expected pension or interest expense, and mild weather weighing on earnings.

• Xcel Energy Inc. (XEL)

  A MISO-related regulated utility benefiting from rising data center demand in MISO.

  Strengths

  Buy rating with a 12-month target price of $93; the report favors its renewable energy and transmission leverage, higher growth, and execution track record.

  Weaknesses

  Needs to continue narrowing the gap between actual ROE and authorized ROE while managing costs.

  Comparison

  Compared with FE's PJM exposure, XEL offers exposure to growth markets in MISO.

  Risks

  Adverse rate case outcomes, litigation, inability to narrow the ROE gap, and cost management issues.

• Duke Energy Corp. (DUK)

  A Southeast regulated utility benefiting from a large load pipeline and non-ISO data center markets.

  Strengths

  Buy rating with a 12-month target price of $145; the report mentions about 15.4GW of high-confidence large-load pipeline and a favorable regulatory environment.

  Weaknesses

  Balance sheet improvement and regulatory progress remain key variables.

  Comparison

  Compared with PJM/MISO/ERCOT names, DUK represents exposure to the largest non-ISO data center market in the Southeast.

  Risks

  Failure to improve the balance sheet, regulatory uncertainty in rate cases, and weaker-than-expected strong load growth.

• Sempra (SRE)

  A beneficiary of transmission and distribution capital investment opportunities related to Texas and ERCOT.

  Strengths

  Buy rating with a 12-month SOTP target price of $109; the report believes it is well positioned to capture Texas opportunities through T&D capital investment.

  Weaknesses

  Management execution, communication, and balance sheet strength are areas of concern.

  Comparison

  Unlike IPPs in ERCOT, SRE is more of a regulated transmission and distribution investment path.

  Risks

  Management execution and communication challenges, weakening balance sheet, and regulatory outcomes falling short of expectations.

• Talen Energy Corp. (TLN)

  An independent power producer benefiting from rising power prices in PJM and ERCOT and incremental data center PPAs.

  Strengths

  Buy rating with a 12-month target price of $499; valuation is based on 10.0x EV/EBITDA and 7.0% FCF yield.

  Weaknesses

  More sensitive to power prices, interconnection approvals, and data center demand elasticity.

  Comparison

  Compared with regulated utilities, TLN has higher earnings sensitivity to rising power prices and PPAs.

  Risks

  Regulatory and interconnection risks, lower-than-expected power prices, and slowing data center demand.

• Vistra Corp. (VST)

  An independent power producer benefiting from PJM capacity prices, power prices, and data center PPAs.

  Strengths

  Buy rating with a 12-month target price of $209; based on 10.5x EV/EBITDA and 7.0% FCF yield.

  Weaknesses

  Has stronger thematic trading characteristics and is sensitive to capacity auctions and power price volatility.

  Comparison

  Along with TLN and NRG, it is part of the IPP beneficiary group.

  Risks

  Uncertainty around PJM capacity auctions, lower-than-expected power prices, and weakening durability of the power demand theme.

• NRG Energy Inc. (NRG)

  An independent power producer benefiting from rising power prices and incremental data center PPAs.

  Strengths

  Buy rating with a 12-month target price of $197; based on 9x EV/EBITDA and 8.0% FCF yield.

  Weaknesses

  Market attention driven by the AI and data center theme may also increase volatility.

  Comparison

  Alongside TLN and VST, it is viewed in the report as one of the IPPs best positioned to benefit from opportunities in PJM and ERCOT.

  Risks

  Headline risk related to the AI and data center theme, lower-than-expected power prices, and uncertainty in PJM capacity auction pricing.

Key data

• 2030 global data center capacity217GWPreviously updated to 168GW in February; the 2025 baseline is 101GW.
• 2025-2030 global net new capacity116GWUsing a rule of thumb of $50 billion/GW, this corresponds to about $5.8-6.0 trillion in capital expenditures.
• Global data center capacity CAGR17%From 2025 to 2030.
• 2030 US data center IT Power supply125GWAbout 76GW higher than in 2025.
• 2030 US data center demand108GW / 947TWhBased on 121GW IT Power, about 1.3 PUE, 70% utilization, and 97% occupancy.
• US total power demand CAGR3.5%Raised from the previous 3.2%, driven jointly by BTM and FTM data center demand.
• 2030 BTM serviceable capacity and power deliveredabout 31GW capacity / about 22GW power deliveredPreviously 20GW capacity / 14GW power delivered.
• 2030 US regional demandPJM 38GW, MISO 17GW, ERCOT 15GWPJM remains the largest data center demand market, while MISO surpasses ERCOT to become the second largest.
• Global data center power demand growth170% growth in 2030 versus 2025The previous expectation was 117%, and more than 60% of the growth comes from the US.
• AI/Cloud player capital expendituresabout $1.214 trillion in 2030EThe table shows total capex for the sample AI/Cloud players rising from $444.975 billion in 2025 to $1,214.176 billion in 2030.
• Data center carbon emissionsabout 370 million tons increase in 2030 versus 2023, about 210% growthThe report estimates this is equivalent to about 1.0% of global energy emissions.
• DLR target price$215Buy rating, unchanged 12-month target price, based on 26x NTM+1Y P/AFFO.

Impact & implications

The investment implication is that data center expansion is not just a technology hardware or cloud capex theme; it is also becoming a cross-sector investment opportunity spanning power, transmission and distribution, natural gas, cooling, REITs, and long-duration leased assets. For data center operators, lower vacancy and scarce new capacity support pricing for renewals and new leases; for utilities and independent power producers, regional power demand, PPAs, capacity auctions, and BTM solutions enhance growth and price elasticity; for the AI infrastructure chain, financing, power supply, components, and community permitting will become the key bottlenecks determining build speed.

Risks

• Community opposition related to electricity prices, power reliability, water resources, noise, waste heat, and AI data center construction could delay project execution.
• Permitting and execution risk remain industry pain points; even with bipartisan support for expedited approvals, NIMBY and local perception risks may still outweigh technological solutions.
• Supply lead times for power equipment, natural gas combined-cycle units, transmission and distribution, and interconnection equipment may constrain construction pace.
• Shortages of skilled labor, especially electricians, may affect execution of transmission/distribution and data center projects.
• If efficiency gains in AI chips, servers, and models outpace demand growth, capex expectations may be revised down.
• Excessively high electricity prices, power reliability, and water constraints may affect acceptance by companies, communities, and regulators.
• Physical environment risks such as heat, humidity, and drought may increase cooling demand, raise PUE, and offset some efficiency gains.
• Independent power producers face risks from lower-than-expected power prices, capacity auction uncertainty, interconnection approvals, and slowing data center demand.
• Rising data center emissions may create regulatory, reputational, and carbon cost pressure.

What to watch

• Whether 451 Research continues to raise its quarterly data center capacity forecasts.
• Whether data center vacancy rates and per-kilowatt lease prices in major US markets continue to compress or rise.
• Whether capex by hyperscalers, private AI labs, SpaceX, and data center platforms continues to exceed consensus.
• The lease terms, pricing, and financing structures of neocloud and GPU-as-a-service deals involving TeraWulf, Hut8, IREN, SpaceX, and others.
• Whether Nvidia's revenue guarantee and revenue-sharing model becomes a replicable financing mechanism for neocloud.
• Progress in data center load realization, interconnection queues, capacity auctions, and PPAs in PJM, MISO, ERCOT, and the Southeast.
• Whether BTM solutions contribute about 20% of US data center demand by 2030 as the report expects.
• Whether constraints in power equipment, natural gas turbines, transmission and distribution construction, and electrician labor improve.
• The actual impact of community opposition, rate cases, water use, and noise disputes on project approvals.
• Changes in low-carbon PPAs, the share of gas-powered supply, and the emissions path of data centers.