There is a version of the grid resilience conversation that the energy industry knows how to have. It centers on infrastructure: transmission investment, generation adequacy, weatherization standards, reserve margins.

It is a conversation about physical systems, and it produces physical solutions. It is also increasingly insufficient.

The U.S. grid is not simply experiencing an infrastructure problem. Rather, extreme weather is challenging the financial assumptions that the industry built around that infrastructure. 

Those are different problems, and they require different responses. The industry has not yet fully grasped the distinction.

A volatility problem, not just a reliability problem

The traditional framing of grid stress focuses on outages - moments when the lights go out. That framing misses half the exposure. For most energy-reliant businesses, the more consequential risk is not a blackout. It is a price spike.

When a winter weather event hit Texas in February 2025, the grid itself held. But localized electricity prices at certain Electric Reliability Council of Texas (ERCOT) settlement points reached nearly $28,000 per megawatt-hour.

For the industrial operators and commercial buyers sitting behind those price nodes, it was a balance-sheet event of the first order, and it happened with no outage, no blackout, no infrastructure failure. The market worked as designed. That is the point.

This is not a Texas anomaly. It is a preview. The North American Electric Reliability Corporation's (NERC) 2025 Long-Term Reliability Assessment projects summer peak demand will surge by 224 GW by 2035 - a figure 69% higher than NERC's own projection just one year earlier.

NERC now flags 13 of 23 North American assessment areas as facing elevated or high resource adequacy risk over the next five years. Meanwhile, confirmed generator retirements will reach 52 GW by 2029, with dispatchable thermal capacity being replaced primarily by weather-dependent resources that cannot be called on demand.

What this produces is a grid with tighter margins and higher sensitivity to more frequent extreme climate events. The system is not failing. It is becoming more volatile.

The forecasting problem no one wants to admit

Here is the uncomfortable reality for grid planners: the models are breaking down, and the cause is not a lack of computing power or talent. The models are trained on historical data from a climate that no longer exists.

During that same February 2025 Texas event, ERCOT's net load forecast error hit a record 17 gigawatts. That was not a model failure in the conventional sense as the model did what it was designed to do: it extrapolated from past performance.

The conditions it was asked to anticipate fell outside any historical distribution it had been trained on. This matters beyond the operational.

Demand forecasting feeds into capital allocation, hedging strategies and rate cases. If the forecasts are structurally biased toward underestimating extreme events, then every downstream decision based on those forecasts inherits that bias.

The cost of that miscalibration shows up eventually: in Houston’s electricity prices which have risen roughly 50% since 2020, in quarterly earnings surprises for industrial operators who thought they were hedged, and in utility rate cases that catch regulators off guard.

The energy market risk tool that is missing

Infrastructure investment is the right long-term response. Grid hardening, transmission expansion, and weatherization are necessary, and the industry is pursuing them.

But they operate on decade-scale timelines while financial exposure is quarterly. That gap between the pace of physical investment and the pace of climate-driven volatility is where a different class of tool becomes essential.

Parametric and weather-linked risk transfer instruments are not new, but their application to energy market risk is underutilized. The core idea is straightforward: Rather than hedging against physical damage after the fact, a company structures a financial instrument that pays out when a defined climate or market trigger is crossed, such as a temperature threshold, a price index level, or a generation shortfall measure.

When those triggers are designed around the company's actual exposure rather than a generic weather index, the result is a hedge that responds to the same events that move the income statement. The basis risk problem that has historically limited weather derivatives is the gap between the trigger and the realized loss.

This narrows considerably when the trigger design is informed by high-resolution climate data and asset-level exposure modeling.

This is where AI changes the calculus. Proprietary foundation models trained on satellite-derived climate observations can produce probability distributions over extreme events that backward-looking reanalysis products simply cannot generate.

Better distributions mean better-calibrated triggers, which means instruments that perform when they are supposed to. For a utility or large energy buyer, that is the difference between a hedge and a hope.

An integrated resilience model

The organizations that will navigate the next decade most effectively are those that refuse to segregate these problems. Physical resilience, forecasting accuracy and financial hedging are not separate workstreams. They are interdependent layers of the same risk management challenge.

Optimizing one without the others leaves compounding exposure in the gaps.

The grid will face more stress. Climate variability is accelerating faster than the consensus models anticipated even five years ago.

The demand side is adding pressure from many directions that weren’t previously forecasted, such as AI data centers, electrification, and industrial growth. What is within the industry's control is whether the financial exposure that follows from that volatility is managed deliberately or absorbed by surprise.

The tools exist. The data exists. What remains is the willingness to treat climate risk as a financial risk, not just an engineering one, and to build accordingly.