The repricing of compute
- Rémi Paccou & Thomas Epelbaum
- 25 Jun 2026
- 5 min read
Climate change is repricing the economics of data center infrastructure. New research from the Schneider Electric Research Institute, analyzing 8,572 facilities across 120 countries, finds that climate risk erodes $388 billion in asset value from a modeled baseline of roughly $1 trillion. That means physical climate exposure puts 38% of the global data center sector’s value at risk before adaptation is accounted for. For AI-era facilities now being built, proactive adaptation delivers a 30% to 39% risk reduction with positive ROI in every scenario tested.
In January 2026, a winter storm paralyzed Virginia’s Data Center Alley. Wholesale electricity prices surged from $200 to $1,800 per megawatt-hour overnight. The PJM grid approached 147 gigawatts of demand, shattering its previous winter record. The Department of Energy issued emergency orders requiring data centers to prepare backup generation.
It wasn’t an isolated episode. Storm Éowyn knocked out power to more than 1 million customers across Ireland and the U.K., disrupting pan-European cloud workloads. A major typhoon in Asia disrupted power systems across a dense digital infrastructure corridor. In the Gulf, temperatures exceeding 50 degrees Celsius pushed cooling systems to their design limits.
The pattern is clear: Physical climate hazards now impose direct costs on the infrastructure that runs the digital economy. And yet the sector that hosts the computational machinery used for climate projections, stores the data behind Task Force on Climate-related Financial Disclosures (TCFD) and Corporate Sustainability Reporting Directive (CSRD) disclosures, and prices risk for every other industry can’t price its own.
$388B
Climate-related value at risk across a modeled baseline of roughly $1T in asset value
×2.6
Physical risk amplification per GW on AI-era infrastructure versus legacy facilities
30–39%
Reduction in climate exposure from proactive adaptation, with positive returns under every scenario tested
Related
Download the visual summary: The climate cost of computingView a visual summary of the five risk channels, regional exposure patterns, and adaptation business case behind the report.
Our latest research from the Schneider Electric Research Institute puts a number on what capital markets have been ignoring. Across 8,572 geocoded facilities totaling 90.8 gigawatts in 120 countries, climate risk erodes $388 billion in asset value — 38% of the total. No current valuation framework captures this figure, and no disclosure regime requires it.
The analysis uses a Climate Value at Risk model, or ClimVaR, that translates physical hazards and transition risks into discounted cash-flow impacts across five economic channels: cooling stress, carbon pricing, business interruption, physical damage, and heat productivity loss. Two analytical tools, neither previously applied to this sector, sharpen the picture. Shapley decomposition, from cooperative game theory, attributes each channel’s contribution to specific hazards. Leontief input-output analysis traces business interruption through upstream supply chains.
The results reframe how we think about data center resilience. Business interruption leads at 31% of total ClimVaR, but 91% of it propagates through supply chains rather than direct facility disruption — a ratio of 11-to-1 between indirect and direct exposure. An operator monitoring only on-site hazards misses the primary risk vector.
The global headline conceals a decisive fact. Europe and China reach ClimVaR intensities of 53% to 56%, while the U.S. stands at 28%. That twofold spread can’t be explained by hazard maps alone. It reflects the interaction between regional climate conditions, electricity systems, industrial dependencies, and carbon policy regimes.
Each region carries a distinct risk fingerprint. In the U.S., business interruption fills nearly half the exposure. Europe shows a different signature: $45 billion in carbon costs alone, shaped by the European Union Emissions Trading System. China carries a triple threat of comparable magnitude across carbon, physical damage, and business interruption, so intervention on any single channel leaves roughly two-thirds of the exposure intact.
The implication is direct: A uniform global adaptation strategy systematically underinvests in the dominant risk channel. The same dollar doesn’t buy the same protection in each geography.
The AI build-out carries a materially different risk profile from existing installed base. On legacy facilities, carbon costs represent 28% of total ClimVaR and physical channels 72%. On AI-era facilities, physical channels account for 93% to 94%, while carbon drops to 6%.
Three mechanisms drive this shift. Geographic concentration: AI facilities cluster in regions combining cheap power with fast permitting — Texas, Virginia, parts of China and Southeast Asia — the same regions with elevated physical climate exposure. Power density: AI racks at 30 to more than 100 kilowatts generate more waste heat per square meter, so a cooling failure at a 60-kilowatt-per-rack AI cluster triggers thermal shutdown within minutes. Supply chain criticality: AI training runs are noninterruptible workloads where a single disruption can invalidate weeks of compute.
Net-zero commitments address the carbon component. But on AI infrastructure, that’s less than 10% of the exposure. Physical resilience, not energy sourcing, is the binding constraint.
Proactive adaptation, including power purchase agreements (PPAs), liquid cooling, microgrids, and climate-informed siting, reduces aggregate ClimVaR by 30% to 39%, with positive net present value under every scenario tested. All 10 adaptation actions measured in the study fall below the break-even line. The cheapest interventions, such as signing a PPA at the design stage, cost 4 cents per dollar of risk saved. The most expensive lever still costs 65 cents per dollar of ClimVaR reduced.
But the bigger message is not that adaptation eliminates the problem. Even after proactive measures, a structural floor remains: 61% of climate-related value at risk is left untouched. For operators with large installed bases, selective adaptation is necessary but insufficient. It can compress the highest-value risk channels, delay avoidable losses, and improve capital allocation, but it cannot erase exposure embedded in geography, grid dependency, legacy facility design, and regional climate trajectories.
The selective strategy, one targeted action per facility on the highest-Shapley-value channel, captures 56% of proactive adaptation’s total risk reduction for 15% to 20% of capital expenditure. It is the low-regret floor. For operators with large installed bases, the full proactive approach is necessary to address the 60% to 70% of capacity that selective measures leave untouched.
Carbon is the most responsive channel, compressing by 63% through power purchase agreements. But the residual 61% of ClimVaR that proactive adaptation cannot eliminate reflects a structural reality: extreme physical events, semiconductor fabrication concentrated in Taiwan, power transformer lead times of 12 to 24 months, and thermodynamic limits on cooling in the hottest climates cannot be engineered away at the facility level.
Between 40% and 60% of 2035 global data center capacity will be built in the next five years. Once commissioned, these facilities can’t easily relocate. Every siting decision made now embeds a quarter-century bet on climate trajectories, or a quarter-century of climate resilience. The economics of that choice are asymmetric: In the U.S., every dollar invested in climate resilience avoids up to $13 in losses.
The adaptation window is narrowing. Every year of delay converts a greenfield opportunity, where resilience is a line in the design brief, into a future retrofit, where it becomes a construction project with higher cost, more downtime, and narrower technical options. The repricing of compute has already begun. The question is whether capital allocation catches up before the design window closes.
Download the full report, “The Repricing of Compute: The Economics of Climate Risk and Resilience for AI-Era Capital,” to explore the full analysis behind the $388 billion climate-related value at risk facing global data center infrastructure.
Latest from the Schneider Electric Research Institute
AI is electrification
AI-Powered microgrids: Optimizing the balance
The barriers and drivers to decarbonization of the building sector in the United States: Qualitative insights from Boston and Phoenix
Time to power: A generational cycle of modernization in the US
Need help?
Product Selector
Quickly and easily find the right products and accessories for your applications.
Get a Quote
Start your sales inquiry online and an expert will connect with you.
Where to buy?
Easily find the nearest Schneider Electric distributor in your location.
Help Center
Find support resources for all your needs, in one place.