ERCOT's data centre build-out is a grid-stability risk, not just a capacity one
The market is pricing demand growth; the harder problem is load that can vanish without warning during the very events it helps cause.
ERCOT officials have been explicit about a risk that sits alongside the much-discussed demand surge: large computational loads that trip off the grid unexpectedly during voltage or frequency events, potentially compounding the very stress that triggered their disconnection. The grid operator flagged in early June 2026 (2026-06-02) that it is "concerned that the tripping of large loads during voltage or frequency events presents a threat to grid reliability," citing observed behaviour rather than theoretical modelling.1
That observation derives weight from what Virginia's grid experienced in July 2024. Roughly 1,500 MW of mostly data centre demand dropped off unexpectedly — the electricity equivalent of power for 375,000 Texas homes at peak. The disconnection caused substations to run at 60.3 hertz instead of the standard 60, tripping protective breakers. One such event in a grid already stretched thin can initiate cascades.1
The consensus framing around North American winter reliability concentrates on the upside of the demand trajectory. ERCOT estimates data centre demand could reach 228 gigawatts by 2032, against the grid's all-time peak of 85.5 GW set in 2023. That projection drives concerns about whether generation additions can keep pace, and it underpins the bearish reliability view.1
But a less-examined implication runs the other way. NERC's Summer Reliability Assessment, issued in early June 2026 (2026-06-03), revised ERCOT's temperature-adjusted demand forecast down by 1.9 GW, or 2.3 percent, largely because modellers updated their assumptions to reflect how large computational loads actually behave at peak — they pull back. The revision implies earlier projections had overstated reliable demand, and that some generation capacity built to cover a peak that does not materialise in the same form during winter cold snaps may represent mispriced risk.3
The second signal the market may be discounting concerns gas supply. NERC's 2025-2026 Winter Reliability Assessment noted that working natural gas storage inventories were approximately 5% above the five-year average entering the heating season, and the EIA anticipated a modest demand slowdown heading into winter. NYMEX Henry Hub front-month contracts traded at $3.15 as of early Wednesday (2026-06-24). A 5% storage surplus is not negligible; it widens the buffer against a cold snap before marginal gas supply comes under pressure, softening the near-term case for elevated winter power prices.2
The bearish case on winter reliability tends to roll these data points into a single headline risk: demand up, supply constrained, grid vulnerable. NERC's own assessment materials are more calibrated. The persistent structural gap is in natural gas infrastructure weatherisation — five winters after Uri, consistent standards for weather-related emergency preparation at critical gas facilities remain absent across states.2 That gap means a storage surplus can erode faster than expected if a prolonged Arctic event exposes equipment failures, limiting the practical value of the 5% cushion. FERC and NERC jointly reviewed system performance during the January 2025 Arctic weather event; whether that review translated into mandatory winterisation action at gas producers and pipelines is the variable that determines how durable the storage buffer is.
The contrarian read is not that winter is safe. It is that the risk is mischaracterised. The grid-stability threat from data centre load is bilateral rather than unidirectional: it adds demand when growing, but it can also vanish suddenly at exactly the moment grid operators need it to be predictable. ERCOT's concern about load tripping at 60-plus hertz is a distinct problem from raw capacity shortage, and the tools for managing it — revised emergency operating procedures, better real-time visibility into large load disconnection logic — are not the same as building more gas peakers or transmission.1
Watch the FERC-NERC January 2025 performance review output and ERCOT's winter operating plans for explicit treatment of large-load frequency response. If that review documents continued weatherisation failures at gas facilities, the storage surplus shrinks faster than the headline 5% suggests when temperatures drop. If ERCOT embeds the updated large-load behavioural modelling — as the 1.9 GW summer revision implies — into winter planning assumptions, the effective peak risk may be lower than 2032 demand projections currently suggest. The 228 GW trajectory matters; how much of it reliably stays connected during a grid stress event matters more.3