The Flexibility Bargain: How Data Centers Trade Curtailment for Faster Power

The Queue That Broke the Old Model

Grid interconnection used to be a first-come, first-served waiting line: a developer applied, engineers studied the request, and the project got a place in a queue that moved at the pace of transmission construction — years, not months. That model has collapsed under the weight of AI-driven demand. In ERCOT’s own filings, 198 GW of large-load capacity applied for interconnection in the first quarter of 2026 alone, with another 86 GW of new requests under review — a figure roughly equal to ERCOT’s entire current peak demand. By November 2025, ERCOT’s large-load queue had already reached 226 GW, up from just 63 GW at the end of 2024 — nearly quadrupling in a single year, with roughly 77% of that volume made up of data centers targeting a 2030 connection date.

Texas is the extreme case, but not an outlier. Nationally, U.S. data center power demand is projected to jump from 31 GW in 2025 to 66 GW by 2027, with data centers’ share of summer peak demand more than doubling from 4.1% to 8.5% over the same window — and some projections put data centers at 17% of total U.S. electricity consumption by 2030. Utilities have already made 187 GW in data center power commitments, an amount that the actual pace of new grid capacity additions cannot realistically match. The mismatch between promised megawatts and buildable megawatts is what forced regulators to stop treating interconnection as a queue problem and start treating it as a flexibility problem.

What FERC Actually Ordered

On June 18, 2026, the Federal Energy Regulatory Commission issued six tailored “show cause” orders to CAISO, ISO New England, MISO, NYISO, PJM, and the Southwest Power Pool, demanding that each grid operator justify — or fix — its large-load interconnection rules. The action traces back to an October 23, 2025 letter from DOE Secretary Chris Wright directing FERC to accelerate data center and advanced-manufacturing interconnection nationwide. FERC’s proceeding drew more than 3,500 pages of public comment before the Commission acted roughly eight months later.

Each RTO and its transmission owners now have 60 days — until August 17, 2026 — to either show cause why their existing tariffs remain “just and reasonable” without reform, or propose specific tariff changes that address five issues FERC flagged: faster transmission-service application and study processes, prevention of cost shifts onto existing ratepayers, new transmission service categories for flexible large loads, co-location and behind-the-meter generation policy, and a dedicated study track for facilities serving “electrically proximate” loads. A single 90-day extension is available, but only if requested within the first 45 days. FERC Commissioner David LaCerte was blunt about what happens if operators stall: “If RTOs fail to address concerns identified by FERC, the agency will dictate the solutions.” Roughly one-third of Americans live outside RTO/ISO footprints and won’t be directly touched by these orders — a reminder that this reform, however aggressive, is still regional rather than national in reach.

The Mechanics of the Flexibility Trade

The core insight behind FERC’s orders is that not all data-center load behaves the same, and treating it as monolithic “firm power” wastes grid capacity that doesn’t need to exist. FERC itself put it plainly: “Load flexibility can avoid inefficient and costly transmission system build-out.” The Electric Power Research Institute’s DCFlex initiative has formalized this into a five-tier FlexMosaic framework (Classes A through E) that lets a data center declare how much and how fast it can shed load — from brief automated curtailments to sustained multi-hour throttling. EPRI’s own demonstrations found AI training and inference workloads can flex 18% to 55% of peak draw without breaching service targets, and in some pilots achieved up to 40% sustained flexibility.

That flexibility has real system-wide value. Utility Dive’s reporting notes that a 1-2% reduction in aggregate data center peak demand can lower electricity rates for all customers by 0.5% to 2.8% — flexible load isn’t just a favor to hyperscalers, it’s a rate-relief tool for everyone else on the grid. The Brattle Group’s modeling goes further, estimating that widespread adoption of flexible data-center loads could deliver more than $110 billion in aggregate customer savings. Utilities are testing the mechanics now: Silicon Valley Power is piloting bidirectional, real-time communication with an NVIDIA-operated facility to verify curtailment signals rather than trust self-reporting, and PJM has shifted from processing interconnection requests one at a time to studying them in geographic clusters — a change that mirrors how flexible-load tariffs bundle risk across many data centers rather than underwriting each one individually. The unresolved tension, as Silicon Valley Power’s COO has noted, is control: utilities want the ability to physically cut load at the breaker, while data-center operators want to retain enough operational autonomy to protect service-level agreements. That negotiation — not the interconnection study itself — is now the actual bottleneck.

What Algeria’s Energy and Data-Center Planners Should Do About It

1. Write curtailable-load tariffs into Sonelgaz’s interconnection process before the first hyperscale request arrives

Algeria doesn’t have a queue like ERCOT’s yet — that is an opening, not evidence of inaction. Sonelgaz can adopt FERC’s core innovation now, before demand materializes: a standing tariff class for large loads that accept curtailment during system stress in exchange for a faster, cheaper interconnection path than firm-power requests get. The economics already favor this. The Brattle Group found flexible load could unlock over $110 billion in U.S. system-wide savings; on a smaller grid, the marginal value of avoiding a new peaker plant per connected megawatt is proportionally higher, not lower.

2. Build the metering and verification layer CREG will need to enforce it

A flexible-load tariff is only as credible as the ability to verify curtailment in real time. EPRI built its five-tier FlexMosaic framework precisely because “flexible” without instrumentation is an unenforceable promise. Algeria’s electricity regulator should specify a minimum telemetry standard — sub-15-minute reporting and a remote curtailment signal — as a condition of any large-load tariff, modeled on the bidirectional pilot Silicon Valley Power is running with an NVIDIA facility, rather than accepting operator self-reporting at face value.

3. Target workloads that are structurally curtailment-tolerant first

AI training and batch inference sit among the most flexibility-tolerant workloads in computing — EPRI’s own data shows 18% to 55% of peak draw can shift without missing service targets. Telecom and cloud operators building edge or AI infrastructure in Algeria are the natural first candidates for a flexible-tariff pilot, because back-office and training workloads can absorb curtailment without the latency penalties that would hit real-time voice or transaction-processing systems.

4. Push for a fixed regulatory review clock, not open-ended study

FERC’s show-cause orders gave six grid operators a hard 60-day deadline, extendable only once and only if requested within the first 45 days. Algeria’s advantage in building this rulebook from a smaller base is that it can copy the deadline discipline directly: a fixed review window for large-load tariff applications removes the single biggest driver behind ERCOT’s queue explosion — years-long uncertainty that pushes developers to over-apply for capacity “just in case,” inflating the queue far beyond what will ever get built.

Where This Fits in 2026’s Grid Economics

The FERC orders mark a shift in how grid capacity itself is allocated: from a static queue position to a negotiated attribute of the load. That reframing extends well beyond U.S. data centers. Any country courting AI infrastructure investment — including Algeria, as it weighs incentives for hyperscale and edge compute — will eventually face the same choice FERC just forced onto six grid operators: build years of new transmission to serve peak-only firm-power requests, or unlock existing capacity by pricing and enforcing flexibility.

The risk cuts both ways. If flexibility becomes the default expectation, utilities and regulators could lean on it to justify underbuilding transmission altogether, quietly shifting reliability risk onto the data centers that accepted curtailable terms — a bet that works until a grid-stress event coincides with a training run a customer can’t actually afford to interrupt. FERC’s own five-issue framework anticipates this by requiring cost-shift protections alongside the flexibility provisions, not instead of them. That pairing — flexibility as a tool for speed, paired with guardrails against quietly deferred infrastructure spending — is the template worth watching, whether the grid in question is ERCOT’s or Sonelgaz’s.

Frequently Asked Questions

Sources & Further Reading

• Data centers negotiate flexibility for speed to power — Utility Dive
• 6 takeaways from FERC’s data center interconnection decision — Utility Dive
• Solving Interconnection Bottlenecks With Data Center Load Flexibility — Renewable Energy World
• FERC Launches Aggressive Targeted Action to Speed Large Load Integration — FERC.gov
• ERCOT’s large load queue has nearly quadrupled in a single year — Latitude Media
• Can US Interconnection Queues Survive Data Center-Driven Load Growth? — Ascend Analytics