01 / XMIGHT POM MANAGER

Facility-level layer that sets the power envelope for each rack from grid conditions and site policy, and records per-rack event evidence for the site's compliance reporting.

INTERFACES

  • Redfish API — telemetry ingestion
  • Site EMS / point-of-interconnection data — grid state and policy
  • POM interface — rack envelope to each Module

Interface contributed to OCP Rack & Power as POM

02 / XMIGHT POM MODULE

Rack-mounted module that coordinates BBU discharge and PSU output inside the ORv3 power shelf. Installs in the PMI-class slot with its own control plane, and keeps operating autonomously if the facility layer is unreachable.

INTERFACES

  • ORv3 power shelf, PMI-class slot
  • SMBus + PMBus control
  • Real-time BBU + PSU coordination

Implements POM — in review, OCP Rack & Power

Diagram of an ORv3 rack: POM modules in the power shelf and the BBU shelf coordinate PSUs and BBUs as one response to a grid event.
POM adds coordination, not energy. In ORv3 that energy sits in rack BBUs; in Mt Diablo architectures it extends to BBU/CBU resources in the sidecar power rack.

Who controls what

  • The Manager decides the envelope.
  • The Module executes locally.
  • Fast protection never depends on the network or the facility controller.
  • Compliance is measured for the whole site at its grid connection point.

CONTROL LOOP

How XMight's POM implementation responds

XMight POM Control-Loop Sequence
Phase 1 Rack · PSU · BBU Telemetry ingested via Redfish + PMBus
Phase 2 POM Manager Envelope and setpoints determined
Phase 3 POM Module Setpoints issued to BBU + PSU
Phase 4 Grid interface Site sees a shaped, predictable rack load
Continuous POM Manager Per-rack event evidence logged

ARCHITECTURE

Set the envelope, coordinate the rack, prove it.

XMight's POM implementation sits between the data center EMS and the ORv3 rack. The diagram below shows POM (Power Orchestration Module) — XMight's open contribution to OCP Rack & Power — and the two-layer architecture of the XMight POM Module and POM Manager.

Built on ORv3 / ORW racks today. POM is topology-agnostic and designed for Mt Diablo ±400 VDC and NVIDIA 800 VDC sidecar architectures, where one power rack feeds several IT racks and recovery concentrates on the power rack.

DATA CENTER EMSFleet scheduler · workload orchestratorPOWER GRIDDemand · capacity · frequency schedulerXMIGHT POM MANAGER — FACILITY LAYERFacility-level envelope and policyVRT POLICYVoltage ride-through envelope enforcementPFAPR RECOVERY CONTROLPost-fault active power recovery slopeCONTROL PLANEXMIGHT POM MODULE — IN THE POWER SHELFRack-mounted power orchestration moduleVRT EDGE POLICYLocal VRT enforcementBBU/PSU ORCHESTRATORDischarge + output shapingLOCAL EVENT DETECTIONAutonomous rack-level responseTELEMETRY AGGREGATORFleet data collectionAC INPUTGrid powerORv3 RACKACPOWER SHELFAC → DC conversionPOM48V DC48V DC BUSBBU SHELFStandard backupCharge ⇄ dischargePOMBBU — VRT/FRT/PFAPRGrid compliance eventsCharge ⇄ dischargePOMAI SERVER ×6+GPU compute load
POM (Power Orchestration Module) — XMight's open contribution to OCP Rack & Power
THE NUMBER THAT MATTERS

AI racks are provisioned with headroom for transient peaks — compute capacity held back. Coordinated BBU discharge absorbs the spikes instead, so the reserve can shrink: XMight's OCP 2025 Active BBU paper models a 25–35% lower peak-to-average ratio. Throttling isn't eliminated; it's made smaller and continuous.

Source: XMight, The Active BBU: Dynamic Power Orchestration for Stable and Efficient ORv3 AI Racks, OCP Global Summit 2025 (Table I).

VOLTAGE EVENT RESPONSE

What happens to an ORv3 rack — with and without POM.

Simulated voltage events, four traces each. Left: conventional ORv3 rack power. Right: the same event with the XMight POM Module coordinating BBU discharge and PSU output.

WITHOUT POM
WITH POM
Scenario A — 0.5 p.u. sag, 500 ms · Without POMSIMULATED · LOAD DROPS · RECOVERY ≫ 2 s100%50%0%Vac (p.u.)100%0%PSU pwr100%0%BBU pwr100%0%48V busbar0 mssag begins≈3 msPFC trips500 msAC recovers2,500 msPFAPR limitRECOVERY > 2 sPFC trips → grid seesload disconnectPSU dead — restart 6–17 s≫ PFAPR 2 s budget
Scenario A — 0.5 p.u. sag, 500 ms · With POMSIMULATED · BUSBAR HELD · RECOVERY < 2 s100%50%0%Vac (p.u.)100%0%PSU pwr100%0%BBU pwr100%0%48V busbar0 mssag begins3–27 msPOM triggers BBU500 msAC recovers2,500 msPFAPR limitRECOVERY < 2 sPOM detects → BBUdischarges → busbar heldCoordinated PSU rampinside the 2 s window
Scenario B — AC loss (< 0.35 p.u.), 150 ms · Without POMNO KEEP-ALIVE · PSU CONTROLLER DIES100%35%0%Vac (p.u.)100%0%PSU pwr100%0%BBU pwr100%0%48V busbar0 msAC fully lost≈3 msPFC trips150 msAC returns2,150 msPFAPR limitRECOVERY > 2 sPFC trips + controllerloses powerCold-boot + 0–5.5 s delay≫ PFAPR 2 s window
Scenario B — AC loss (< 0.35 p.u.), 150 ms · With POMPOM KEEP-ALIVE · CONTROLLER SURVIVES OUTAGE100%35%0%Vac (p.u.)100%0%PSU pwr100%0%BBU pwr100%0%48V busbar0 msAC fully lost3–22 msPOM sustains busbar150 msAC returns2,150 msPFAPR limitRECOVERY < 2 sPOM + BBU keep busbar alivecontroller stays onlineCoordinated recoverywithin PFAPR 2 s budget

Without POM, the PFC trips within milliseconds and a PSU cold start takes 6–17 seconds (OCP DCF LVDC white paper v1.1) — far beyond the 2-second PFAPR window. With POM, the XMight POM Module detects the event locally, coordinates BBU discharge to hold the 48 V busbar, and the PSUs ramp back inside the 2-second window in simulation. Same distributed-power architecture, with an added coordination path.

Simulated for illustration; not a measurement of a specific product.

POWER FLOW

From GPU transient to grid-visible load.

XMight's POM implementation is designed to absorb the transient inside the rack, before it reaches the site's grid connection.

GPU compute
AI workload begins inference or training step — GPU draw spikes within milliseconds.
XMight POM Manager
The Manager sets the rack envelope from grid state and site policy; the Module watches shelf telemetry locally and acts within it.
XMight POM Module
The Module issues setpoints to BBU and PSU simultaneously — shaping the aggregate draw seen at the rack's grid connection point.
Grid interface
The site sees a shaped, predictable rack load — the rack's contribution to the ride-through and recovery response measured at the point of interconnection.

INDUSTRY ANALYSIS

Four gaps that close-the-loop control must address.

Mapped against current OCP ORv3 specifications. Each gap belongs to a different leaf-spec owner — none can be closed by a single vendor or a single spec change. POM defines the coordination layer; compatible PSU, BBU and PMI implementations expose the capabilities it needs.

VRT GAP
01

PSU PFC low-voltage behaviour

ORv3 PSU operates 180–305 V AC; specification permits derating between 180–198 V. Below 180 V (~0.65 p.u. of 277 V), PFC shuts off in approximately 20 ms with 20 ms holdup. Under ERCOT NOGRR 282 the load must already ride through deeper sags, and from 1 January 2028 it must keep consuming power down to 0.50 p.u.

PFAPR GAP
02

PSU controller power-on delay

PSU controller loses power on AC drop; on restart the ORv3 PSU specification adds a randomized 0–5.5 s power-on delay for inrush staggering, on top of a cold-boot of several seconds. ERCOT's PFAPR window is 2 seconds from voltage recovery — current behaviour cannot meet it.

COORDINATED SAG GAP
03

BBU trigger condition

BBU Module 1.4 §4.5 specifies discharge trigger at busbar < 48.5 V for 2 ms. This covers full AC loss but does not provide coordinated voltage-sag support.

PMI GAP
04

PMI interface scope

PMI Specification 1.0 §2 defines pass-through monitoring only. No write commands, no open rack-level read-write coordination path is currently defined.

Closing the four gaps takes the POM coordination layer plus compatible PSU, BBU and PMI behaviour — which is why POM is proposed as an open OCP interface.

OCP INTEGRATION

POM: the open interface the XMight POM Module implements.

Our OCP contribution defines POM — a new module class in the PMI-class slot with its own control plane — together with the coordinated PSU, BBU and PMI amendments that grid ride-through needs: PSU input envelope, controller keep-alive and fast resume; BBU sag-triggered discharge; a read-write coordination path in PMI. The table maps each leaf spec to the requirement it must meet.

Specification Version ERCOT requirement Status
Open Rack V3 Base 1.0
VRTFRT
Gap identified
Power Shelf 1.0
VRTPFAPR
Gap identified
48V PSU 1.0
VRTPFAPRFRT
Gap identified
PMI 1.0
VRTPFAPR
Gap identified
BBU Shelf 1.1
VRTPFAPR
Gap identified
BBU Module 1.4
VRTPFAPR
Gap identified
Modbus Register Map 0.73
PFAPR
Under review
OUR CONTRIBUTION TO OCP

The Active BBU: Dynamic Power Orchestration for Stable and Efficient ORv3 AI Racks

Presented at OCP Global Summit 2025

This paper describes the Active BBU architecture — the technical foundation of POM — and its role in rack-level post-fault recovery while reclaiming compute headroom traditionally reserved for transient buffering.