The Electric Reliability Council of Texas (ERCOT) gave the market another concrete reason to stop pretending the grid can absorb unlimited hyperscale load growth on top of an already strained generation mix.
In a May 21st report, ERCOT disclosed that multiple clusters of proposed data centers and crypto facilities failed voltage ride-through testing. When subjected to simulated routine voltage disturbances, such as the kind caused by transmission faults, capacitor switching, or equipment issues, four groups of these large users simply disconnected. Models showed each group capable of removing more than 5,000 MW of demand in one event.
“Those abrupt drops in demand were equivalent to the electricity consumption of a large city such as Boston”
In a real-world fault on the Texas grid, those facilities would not ride through the sag and remain online like traditional industrial customers. Their protection systems would trip them offline to protect servers and mining rigs.
Only Texas is aware of the power demand tsunami that is coming. The US is woefully unprepared for the coming explosion in electricity demand pic.twitter.com/9wfgBbS6D8
— zerohedge (@zerohedge) December 13, 2024
The instantaneous loss of thousands of megawatts of demand creates an immediate generation surplus. Frequency rises sharply. Other units can trip on over-frequency protection or be forced into abnormal operation. In tight reserve conditions or during summer peak, the event does not stay localized. It becomes a system stress event.
ERCOT has already recorded at least 26 such disconnection events involving data centers or crypto operations since 2023. The operator is now reviewing roughly 20 GW of large-customer applications, including several gigawatts slated to energize before July. The board has elevated voltage ride-through performance to a top priority precisely because the scale of these new loads makes the old assumptions obsolete.
This is such a fascinating graph. A frequency drop of 0.15Hz was enough to take down Spain and Portugal. pic.twitter.com/tZ1OrITtMU
— andi (twocents.com) (@Nexuist) April 28, 2025
This is the demand-side mirror image of what happened in Spain on April 28, 2025. As we covered extensively at the time, the Iberian blackout was not a simple “too much solar” story. ENTSO-E’s final report pointed to gaps in voltage and reactive power control, differences in how generators responded to voltage swings, and rapid output reductions and disconnections that cascaded across the peninsula.
Many renewable resources were operating in fixed power-factor modes that did not provide dynamic voltage support when the system needed it most. The result was fast voltage increases followed by widespread generator tripping. Natural gas units ultimately helped stabilize the system in the recovery phase, a point we noted when the “net-zero death” narrative was being walked back in real time.
U.S. officials have already flagged the risk of Spain-style events on this side of the Atlantic. Now ERCOT is stress-testing the other half of the equation: what happens when the new hyperscale loads themselves become the trip risk during otherwise manageable disturbances.
We have documented for years how Texas electricity demand could more than quadruple under data-center and crypto growth scenarios, how PJM is scrambling to find 15 GW of new supply for its own data-center alley, and how the largest U.S. grids are operating with minimal spare capacity while aging infrastructure and retiring dispatchable plants reduce headroom. The common thread is not ideology about any single fuel.
It’s physics.
Inverter-based resources and large blocks of sensitive electronic load both behave differently from the synchronous machines the grid was designed around. They offer less inherent inertia and different voltage and frequency response characteristics. When protection settings on either the generation or load side are not aligned with system needs, routine disturbances can escalate.
Before the outage hit, Spain was running its grid with very little dispatchable spinning generation, and therefore no much inertia.
Solar PV/thermal + wind: ~78%
Nuclear: 11.5%
Co-generation: 5%
Gas-fired: ~3% (less than 1GW)Snapshot at 12.30pm local time (outage was 12.35pm) pic.twitter.com/fF7FiIB6UD
— Javier Blas (@JavierBlas) April 28, 2025
That is why the push for new nuclear, new gas-fired capacity with fast-start and flexible capability, and retention of existing dispatchable resources where they still make economic sense is not optional window dressing. It is the engineering requirement for keeping the lights on while AI infrastructure scales.
Renewables can and will continue to grow, but they bring additional control challenges that the current grid architecture and market rules were never sized to handle at this speed and volume.
The Spain event demonstrated the supply-side version. ERCOT’s latest tests are showing the demand-side version. Both point to the same conclusion: you cannot substitute megawatts of intermittent or highly sensitive capacity for the stabilizing attributes that nuclear, gas, and coal plants provide at scale.
Tyler Durden
Mon, 06/08/2026 – 04:15
