Climate risk and the value of data centers: what 38% at risk means for valuation, capital allocation, and resilience planning

Climate-related costs and losses could erode 38% of the value of the world’s existing data centers over the next 25 years. That is USD 388 billion of a USD 1,012 billion asset base. The estimate comes from The Repricing of Compute, a June 2026 study published by the Schneider Electric Research Institute and co-authored by SE Advisory Services, Schneider Electric’s global consulting practice, which modeled 8,572 facilities across 120 countries. Carbon costs, mainly linked to electricity use, account for 28% of that exposure. The remaining 72% comes from physical risks: business interruption, physical damage, and higher cooling costs. Most valuation and disclosure frameworks leave those physical risks unquantified.

The imbalance grows with AI capacity. On planned AI facilities, physical risks account for 93% to 94% of modeled loss and carbon costs for 6% to 7%. A net-zero commitment addresses the carbon component and leaves 82% to 88% of AI-era exposure untouched. For anyone building an investment case on renewable electricity procurement alone, that is the size of the blind spot.

GRESB’s Real Estate and Data Center Assessments now ask where climate risk comes from and what it costs. The 2026 Real Estate Assessment scores whether entities identify physical and transition risks and assess their financial impact. The Data Center Assessment covers physical climate risk and resilience alongside energy, emissions and water. Neither can be completed well without knowing which of those sources of loss drives the value at risk at each site.

The Repricing of Compute study covers existing data centers plus planned AI capacity under three growth scenarios, and estimates climate value at risk: how much lower a data center’s value becomes once expected climate-related costs and losses over 25 years are taken into account.

The financial model looks at five ways in which climate change impacts a data center’s finances: higher cooling costs, carbon costs business interruption, physical damage, and heat affecting outdoor workers. It traces business interruption back through suppliers, and weighs adaptation costs against the losses avoided.

Unless stated otherwise, figures assume a high-emissions climate pathway (known as RCP 8.5) and current carbon policies.

More than a third of today’s data center’s value is at risk. The model estimates that climate-related costs and losses would remove USD 388 billion, or 38%, from the USD 1,012 billion value of existing data centers. For a data center worth USD 100 million before climate effects, this means about USD 38 million lost gradually over its lifetime, with about a quarter of it in the next five years. Adding planned AI capacity raises the total to between USD 1.0 trillion and USD 3.7 trillion by 2050, depending on how fast AI grows. Under a net-zero policy pathway, higher carbon prices lift the figure for existing sites to USD845 billion, or 83% of its value.

Most business interruption impacts start outside the site. Business interruption from physical risks is the largest source of modeled loss, at 31% of the total. For every dollar of it caused by hazards at the site, about 11 dollars comes through the supply chain (e.g.,when typhoons disrupt chip plants). At each of the three sites examined in detail, 98% to 99% of it originated upstream. The IEA reports that bottlenecks in energy supply chains and advanced chip manufacturing have tightened over the past year.2 A site can look safe on a flood map and still be exposed through its suppliers and its grid.

Location changes the type of risk, not just the amount. Modeled risk ranges from 28% of value in the United States to 56% in China and the main driver differs: business interruption in the US (44%), carbon costs in Europe (43%), and physical damage in Asia-Pacific (39%). In China, the three drivers are of similar size, so tackling any one leaves about two-thirds of the risk in place. A single global resilience budget can therefore put money in the wrong place. At one modeled French site, river flooding pushes losses above the site’s entire value, a signal for engineering checks rather than a verdict. Separate research by XDI, a physical climate risk analytics firm, also finds that exposure varies widely between locations, even within one country.

New AI facilities carry 2.6 times more physical risk per unit of capacity than existing ones. Their higher value explains part of that; three other factors matter. AI sites cluster in regions chosen for cheap power and fast permits, which are often more exposed to climate hazards. Their racks generate far more heat, so a cooling failure can force a shutdown within minutes. And training runs cannot easily be interrupted: one disruption can waste weeks of computing work.

Resilience pays best when it is designed in. A full adaptation program cuts losses by 30% to 39%, protecting USD 150 billion, with net positive returns in every scenario tested. All 10 measures modeled, among them long-term renewable power contracts, liquid cooling, microgrids, and climate-informed site selection, cost less than the risk they reduce. Costs range from about four cents per dollar of risk reduced to 65 cents. Designing resilience into new builds adds an estimated 15% to 20% to capital costs, against 40% to 60% for retrofits, so timing drives cost. About 61% of the risk remains beyond the reach of site-level action. XDI likewise finds that design changes could cut the number of high-risk data centers by more than two-thirds by 2050, but warns that adaptation alone is not enough without emissions cuts.

Scope decides what an assessment can see. The 2026 Real Estate Assessment asks for a systematic process to identify physical risks, and its definition of longer-term climate stresses includes effects on upstream and downstream suppliers. For data centers, an assessment that stops at the site boundary may miss the main source of business interruption risk. Transition risk indicators matter too: under a net-zero policy pathway, carbon costs become the larger exposure for existing sites.

GRESB’s physical risk impact indicator asks for evidence of financial outcomes, and notes that the most sophisticated assessments address probability and uncertainty. The Data Center Assessment credits current and future vulnerability assessments, adaptation options linked to hazards with their costs and benefits, and an investment case for resilience. A breakdown of what drives risk at each site, with costs options attached, to answers both.

The findings also bear on each stage of investment. At acquisition, sustainability due diligence is already scored, and supplier and grid dependencies deserve a place alongside site hazards. During development, resilience is far cheaper to design in than to add later. In operation, decarbonization plans need a resilience plan alongside them. A pricing question sits behind all three: a portfolio with a disclosed climate value at risk of 38% should not be priced like one at 15%, and no disclosure framework currently lets investors tell them apart.

For hyperscale developers building mostly new capacity, focusing first on new builds captures more than half of the full benefit at the lower cost. Colocation providers and others with large existing portfolios need a fuller program that includes retrofits, because the 60% to 70% of today’s capacity will still be operating in the 2040s.

Which climate future are you planning for? The headline figures use a high-emissions pathway. Some of its assumptions, such as continued high coal use worldwide, now look less plausible, but stronger climate sensitivity or carbon-cycle feedbacks could produce similar impacts at lower emissions. Under a net-zero policy pathway, the risk shifts toward carbon costs instead. Testing more than one scenario shows which decisions hold up.

How far does your risk view reach? Supplier dependencies drive most modeled business interruption risk. The model assumes firms cannot switch suppliers during a long disruption, so owners that can reroute sourcing may face less. Supply chain redundancy and continuity planning are worth assessing alongside site hardening.

What about water? Apart from business interruption caused by drought, the model excludes water stress, as it does litigation and rising electricity prices. All three would add to the total. GRESB treats water use and efficiency as material for data centers, and in the EU operators of data centers of at least 500 kW must now report water, energy, and other indicators each year.

How much weight should one number carry? A peer-reviewed comparison of six physical risk scores found large differences between them, even where methods were similar. XDI, which focuses on physical damage to buildings, classed about 6% of data centers as high risk in 2025. It defines high risk as value at risk above 1% of asset value. Site-level physical risk is only one part of the picture, however, the 38% figure also counts business interruption, cooling, carbon and supplier effects. Portfolio figures use regional benchmarks, so site decisions need site-level data.

What will insurance cost? XDI estimates that data center insurance costs could triple or quadruple by 2050 without decisive mitigation and adaptation. GRESB’s Data Center Assessment already recognizes insurability as part of the investment case for resilience.

• Which climate scenarios and time horizons does the physical risk assessment use?
• Does it cover power grids and key suppliers, or only on-site hazards?
• Are financial impacts estimated for each site, or taken from regional averages?
• Is water stress assessed, and how material could it be?
• How are resilience measures built into new developments at the design stage?
• How do decarbonization and resilience plans fit together?

This article was written by Victoria Naipal, Manager Physical Risks, Omer Farooq, Managing Consultant, and Thomas Epelbaum, Principal Financial aluation. Learn more about Schneider Electric Advisory Services here.

1. Schneider Electric Research Institute and SE Advisory Services: The Repricing of Compute https://www.se.com/ww/en/insights/sustainability/sustainability-research-institute/repricing-compute-economics-climate-risk-resilience-ai-era-capital
2. GRESB (2026) Risk Management, 2026 Real Estate Assessment. GRESB Guides. https://guides.gresb.com/completingassessments/2026-real-estate-assessment/re/man/risk-management
3. SB (2026) GRESB Data Center Assessment. https://www.gresb.com/products/data-center-assessment/
4. Harme, J.N. (2025) ‘Climate risks threaten 27% of global data centres by 2050’, DataCenterNews, 11 July, reporting XDI’s 2025 Global Data Centre Physical Climate Risk and Adaptation Report. Also published by DataCentreNews UK. https://datacenter.news/story/climate-risks-threaten-27-of-global-data-centres-by-2050 and https://datacentrenews.uk/story/climate-risks-threaten-27-of-global-data-centres-by-2050
5. GRESB (2026) 3: Physical Risk Identification, 2026 Real Estate Assessment. GRESB Guides. https://guides.gresb.com/completingassessments/2026-real-estate-assessment/re/man/risk-management/rm6.3-physical-risk-identification
6. GRESB (2026) 4: Physical Risk Impact Assessment, 2026 Real Estate Assessment. GRESB Guides. https://guides.gresb.com/completingassessments/2026-real-estate-assessment/re/man/risk-management/rm6.4-physical-risk-impact-assessment
7. GRESB (2026) Physical Risk & Resilience, Data Center Assessment. GRESB Guides. https://guides.gresb.com/data-center-assessment/2.-completing-gresb-assessment/data-center-assessment/management/physical-risk-and-resilience
8. GRESB (2026) 4: Physical Risk Impact Assessment, 2026 Infrastructure Asset Assessment. GRESB Guides. https://guides.gresb.com/completing-gresb-assessments/completing-the-assessment/2026-asset-assessment/management-component/risk-management/rm4.4-physical-risk-impact-assessment
9. Hausfather, Z. and Peters, G.P. (2020) ‘Emissions: the “business as usual” story is misleading’, Nature, 577, pp. 618-620. https://doi.org/10.1038/d41586-020-00177-3
10. Hausfather, Z. and Peters, G.P. (2020) ‘RCP8.5 is a problematic scenario for near-term emissions’, Proceedings of the National Academy of Sciences, 117(45), pp. 27791-27792, discussing Schwalm, Glendon and Duffy (2020). https://doi.org/10.1073/pnas.2017124117
11. Finnish Government (2024) Reporting from data centres to the European database has started. Press release, 14 October. https://valtioneuvosto.fi/en/-/11120570/reporting-from-data-centres-to-the-european-database-has-started
12. Hain, L.I., Kölbel, J.F. and Leippold, M. (2022) ‘Let’s get physical: Comparing metrics of physical climate risk’, Finance Research Letters, 46, 102406. https://doi.org/10.1016/j.frl.2021.102406
13. Mongabay India (2025) Data centres face physical climate risks, report warns. July. https://india.mongabay.com/short-article/2025/07/data-centres-face-physical-climate-risks-report-warns