Interface Specification Between Hyperscalers and Utility Grid for Disturbance Performance (Version 1.0.0)

Overview

The document responds to a problem created by AI and machine-learning data centres of hundreds of MW connecting to the grid. It cites a simultaneous 1,500 MW load loss in North America in August 2024 and a larger event in the Eastern Interconnection on 22 July 2026, which pushed Bulk Electrical System (BES) frequency and voltage up enough to raise reliability concerns. The specification covers only disturbance performance: low voltage ride-through (LVRT) and post-fault active power recovery (PFAPR).

Past large-load requirements were limited to capacity studies, protection and control coordination, IEEE 1453 flicker and IEEE 519 harmonic limits, a minimum power factor at the point of interconnection (POI), for example 0.98 lagging, and, at some sites, NERC PRC-006-2 load shedding. The paper summarises the focus areas in the ESIG large-load report (voltage and multiple-disturbance ride-through, active power recovery, frequency ride-through, ramp rates, reactive power, phase jump, monitoring, modelling). Only voltage ride-through and recovery are covered now; ramp rates and forced oscillations are deferred.

The proposed table defines zones by POI voltage in per unit. 0.90-1.10 p.u. is continuous normal operation. Between 1.10 and 1.20 constant power is held for 1.0 s, and above 1.20 an instant transfer to backup is allowed. Ride-through is 2.0 s for 0.80-0.90, 0.5 s for 0.50-0.80, 0.25 s for 0.35-0.50 and 0.15 s below 0.35. Below 0.50 p.u. power may fall to any value, even zero. Once voltage returns to 0.90-1.10 p.u., load is restored to 90% of its pre-fault value within 2 seconds. The envelope closely mirrors ERCOT's NOGRR 282 proposal.

The commitment applies only to the computational load, which is typically about 70-90% of total data centre load (the MISO convention). Cooling loads sit behind VFDs, follow an IEEE 1566 envelope and can take minutes rather than seconds to restart; firmware alone will not fix them and hardware redesign or upstream energy storage would be needed. Auxiliary loads below 1% are ignored. True ride-through at zero power would require extra capacitance in UPS and PSU designs and could breach the current limits operators set on recovery.

Compliance routes are rack rectifier/PSU firmware updates (manufacturer-specific, timing unpredictable), in-line double-conversion MV-UPS (feasible in new designs, usually too costly and disruptive as a retrofit) and grid-side dynamic reactive support such as SVCs and E-STATCOMs. The proposed implementation plan grandfathers facilities operating or under executed interconnection agreements at the effective date for their lifetime, applies the rules only to incremental growth at existing sites, and expects later facilities to comply at start of operation. Future work lists frequency ride-through, rate of change of frequency, phase jump, counter-based schemes, ramp rate and forced oscillations.

Key points

1. The specification covers only disturbance performance (LVRT, HVRT, PFAPR); ramp rate, frequency ride-through and phase jump are left to a later version.
2. Normal range is 0.90-1.10 p.u.; 1.10-1.20 p.u. requires 1.0 s ride-through, and above 1.20 p.u. an instant transfer to backup is allowed.
3. Low-voltage ride-through times: 2.0 s for 0.80-0.90 p.u., 0.5 s for 0.50-0.80, 0.25 s for 0.35-0.50 and 0.15 s below 0.35.
4. Between 0.50 and 0.90 p.u. active power continues to be drawn and may be reduced in proportion to voltage; below 0.50 p.u. it may be reduced to any value.
5. When voltage returns to 0.90-1.10 p.u., load must be restored to 90% of pre-fault within 2 seconds (PFAPR).
6. The commitment applies only to computational load (about 70-90% of the total); cooling load and auxiliary load below 1% are outside the commitment.
7. In ERCOT NOGRR 282, cooling load is subtracted from the pre-fault load used to judge PFAPR.
8. Voltage is measured at the POI, line-to-neutral or line-to-line, with detailed implementation left to interconnection and protection coordination.
9. Rapid reclosing after a fault can look to data centre protection like a deeper or longer fault, so standardised transmission reclosing schemes are requested.
10. The proposed plan grandfathers existing and contracted facilities for life and applies the requirements only to incremental volume at existing sites.

Using it in design

• For a new site, write the POI LVRT/HVRT envelope and the PFAPR requirement (90% within 2 s) as design criteria, and settle which voltage measurement (line-to-neutral or line-to-line) governs during protection coordination with the utility.
• Classify computational load (rack PSUs/rectifiers) and cooling load (VFDs) separately; obtain written statements from PSU suppliers on firmware and capacitance options, and from cooling suppliers on the IEEE 1566 envelope and restart time.
• If you are evaluating MV-UPS, note that in-line double conversion can supply stored energy for ride-through, but retrofitting it to an existing facility is often impractical because of space and interconnection access.
• Make relay settings, counter-based trip logic and backup transfer logic a specification item to be tested with the utility so that rapid reclosing sequences are not treated as a deeper fault.
• For expansion of an existing site, check the assumption that only incremental MW must comply against the effective-date and grandfathering rules of the specific grid operator.

Who should read it

Electrical engineers working on grid interconnection, medium-voltage and protection design, and data centre developers and grid operators managing large-load interconnection.

Document contents

1. License
2. Compliance with OCP Tenets
3. Introduction
4. Base Specifications
5. Description
6. Past and Upcoming Utility Requirements
7. Detailed Requirements
8. Voltage Ride-Through and Post Fault Active Power Recovery
9. Proposed specification
10. Additional Discussion (power consumption, voltage measurement, reclosing, cooling and auxiliary loads)
11. Possible mitigations (PSU firmware, MV-UPS, grid-side)
12. Implementation Plan Considerations and Outline
13. Future Work
14. References

Read the document

Publication details

Authors

Mike Della Penna; Nithish Mahalingam; Ashish Upreti; Luz Belmont; Mauricio Cespedes; Sai Somayajula; Slavko Vasilic

Publisher

Open Compute Project Foundation

Year

2026

Document language

English

Source type

Industry body

Access

Open licence — downloadable

Licence

OWFa 0.9 — Open Compute open specification licence; redistribution permitted

Licence evidence

www.opencompute.org

Citation