In 82 seconds, ~60 Virginia data centers dropped 1,500 MW off the grid — not from a blackout. Their own backup logic did it. July 10, 2024. A lightning arrestor failed on a 230 kV line near Fairfax. The grid tried to reclose the line three times. But most data centers run a "three-strike" UPS rule — after three voltage dips in a minute, they stop trusting the grid and transfer fully to backup, and they stay there until a human reconnects them. So the third reclose hit the third voltage dip, and ~60 facilities islanded at once. Grid operators never saw it coming, because no NERC standard requires anyone to disclose that ride-through behavior. They later called it a "byte blackout." Here's why this became a board-level problem: PJM capacity prices went from $28.92/MW-day to $329.17/MW-day in two auction cycles. For a 100 MW facility, that's a jump from roughly $1.1M to $12M a year. NERC is writing the first large-load standards by end of 2026. Virginia is weighing a new rate class for anything over 25 MW, plus a moratorium that didn't pass this session but isn't dead. The operators who survive this aren't the ones with the most megawatts. They're the ones who can prove their facility helps the grid instead of threatening it — demand response when PJM is tight, a documented ride-through profile, a capacity-market position that's managed, not just paid. At today's prices, a 100 MW site that can flex 20% of its load turns roughly $2.4M/year of that obligation back into revenue. The proof exists: a 2025 EPRI pilot cut data center draw 25% for three hours with no hit to AI workloads. The orchestration software to do that on a multi-tenant colocation floor mostly doesn't. That gap — between "energy efficient" and "grid friendly" — is the whole game now. Save this if you run a facility in a constrained grid zone. What's your three-strike setting today — and does your utility know it? #DataCenters #GridFlexibility #DemandResponse #NERC #EnergyInfrastructure