XMight is exhibiting at OCP Global Summit 2026 · San Jose · 12–15 October Meet us →

1 Jun 2026 · Young Liang, Founder & CEO · 2 min readInsights

ESS is the muscle. Coordination is what's missing.

Grid codes define behavior at the point of interconnection. The energy to ride through is already in the rack; what ORv3 halls lack is the layer that makes thousands of devices behave as one auditable load.

ESS is the muscle. Coordination is what's missing.
The grid models one aggregate load at the point of interconnection. What actually responds is thousands of PSUs and BBUs, each on its own local logic.

In May I wrote about ERCOT NOGRR 282 — ride-through below 0.35 p.u., a 2-second recovery window, in force since August 2026. Here's the structural follow-on for AI data center operators.

AI DC power architecture is shifting away from traditional facility-UPS designs. NVIDIA's 2025 800 VDC disclosure puts up to a 30% TCO improvement on the table versus today's 54 V DC-bus racks. OCP-native paths — ORv3 and ORW today, ±400 V Mt Diablo / ±800 V LVDC tomorrow — trade centralized facility UPS for distributed rack-tier power infrastructure. The direction is set: it's why hyperscalers and GPU Neoclouds choose this path.

But NOGRR 282, IESO TR v1.0 and NERC Project 2026-02 define behavior at the Point of Interconnection (POI). Whether the architecture is facility-UPS, facility-BESS or distributed-BBU, the compliance question is the same: ride through, recover within seconds, and prove it.

For OCP-native AIDC, an ESS upgrade — facility-side BESS, sidecar BESS or expanded rack-tier storage — is the obvious compliance muscle. But ESS alone, at any tier, runs into two structural problems:

  1. Oversizing risk. Without rack-level visibility, facility ESS gets sized against worst-case aggregated load — a capex overhead.
  2. Audit-trail gap. Grid operators increasingly need better real-time data, model validation and post-event evidence. Facility-level telemetry shows aggregate MW — but not which rack, PSU, BBU or control layer responded during the event.

And the obligation applies even to facilities with no facility-level battery at all. As the Data Center Coalition noted on the record, ERCOT's assumption that every facility has 1:1 backup energy isn't true — for a growing class of OCP-native facilities, ride-through energy isn't in a central UPS; it's distributed across rack PSUs and BBUs.

Where the coordination layer sits: POM in the ORv3 power shelf and BBU shelf, coordinating existing PSUs and BBUs
The energy is already in the rack. The missing piece is the layer that makes it act as one — and records that it did.

That reframes the compliance problem. The energy can be there — ESS is the muscle. What's missing is coordination: thousands of independent rack devices must behave as one predictable, recoverable, auditable load at the POI.

IEEE's IC25-004 points the same way: it calls for interoperability ensuring "consistent behavior across multi-vendor systems" so operators can rely on "predictable site-level responses."

That rack-to-facility coordination and observability layer is the part the industry hasn't built yet. It's the problem space we work on at XMight Smart Energy.

ESS provides the energy. Rack-level coordination turns distributed energy into verifiable grid behavior.

If you are working on ride-through, PFAPR, ESS sizing or rack-tier power architecture for large computational loads, we would be happy to compare notes.

References

  1. ERCOT NOGRR 282
  2. NERC, Reliability Guideline: Risk Mitigation for Emerging Large Loads (May 2026)
  3. IEEE IC25-004 (January 2026)
  4. NVIDIA, 800 VDC Architecture Will Power the Next Generation of AI Factories (2025)
  5. Data Center Coalition comments on NOGRR 282 (February 2026)