How Aravolta Connects Your Colocation With NVIDIA DSX AI Factory Platform

What are NVIDIA DSX and DSX Exchange.

NVIDIA DSX is NVIDIA's platform for designing, building, and operating AI factories. It covers the full stack, from the grid and the facility to the chips and the software that runs on them, so a site produces more useful AI output from the power and infrastructure it has.

DSX Exchange is the open source event bus within DSX. It carries real-time signals between power, cooling, building management, the grid, and the software that schedules compute, so the whole facility can operate as one system.

Once a facility publishes into DSX Exchange, the rest of the platform can act on that data.

The more of the facility that publishes into DSX Exchange, the more the AI factory can do with it.

Why your site should be DSX Exchange optimized.

GPU clouds run their sites on two numbers: tokens per watt and PUE. NVIDIA frames the AI factory as a power-constrained industrial system where the question is how much AI output each available megawatt can deliver, and DSX is built around that number. DSX MaxLPS moves rack power to where the workload is instead of reserving it for a peak that rarely comes, which NVIDIA projects at up to 40 percent more Rubin GPU capacity in a 100 megawatt AI factory. Liquid cooling at 45 degrees Celsius runs the plant with fewer chiller hours and improves annual PUE. DSX Flex makes the site a grid asset that can shed load within seconds when the utility asks.

None of that runs on GPU telemetry alone. The power software wants live meter data, the thermal agents want cooling headroom, the scheduler wants the grid event. NVIDIA's guidance to operators is to design or retrofit for DSX MaxLPS at the site level, existing facilities included, and DSX Exchange is the bus that carries the site into that loop, exposing seasonal cooling headroom and facility power events the software can act on. A facility that publishes runs as part of the AI factory. One that does not runs beside it.

A site that publishes offers the full stack on day one: dynamic power, cooling headroom, grid response, leak isolation, all fed by the plant. Its PUE and its tokens per watt move together.

That is the level the GPU clouds are building to. NVIDIA's cloud partners are deploying DSX components to bring capacity online faster, and when one of them evaluates your hall, the facility side either already speaks DSX Exchange or becomes an integration project on their critical path. Getting there takes the data you already have, published in the shape DSX Exchange expects. That is what Aravolta does.

Put your site on NVIDIA DSX Exchange in weeks.

Aravolta puts a node in your data center. It scans the network, connects to the building and electrical equipment, and gives operators one API to read from and write to all of it. Aravolta nodes already run in more than 100 locations across eight countries.

For a new build, Aravolta can be the BMS and EPMS layer from day one, so the site publishes into DSX Exchange as the plant comes up. For an existing site, the data already lives across a BMS, an EPMS, controllers, gateways, meters, PDUs, and rack gear, each with its own protocol and naming. The Aravolta Node reads all of it and publishes it into DSX Exchange in the BMS AsyncAPI model, so those sites have an on-ramp too.

Either way, keep the BMS you have and Aravolta publishes alongside it, or let Aravolta fill the BMS role in the DSX Exchange specification. No controller reprogramming, no new gateways, no added work for your team. The same node runs your own dashboards, alerts, and controls through the Aravolta API, so it pays off beyond the AI tenant.

ModbusMeters, PDUs, RPPsBACnetCRAH, chillers, CDUsSNMPUPS, ATS, sensorsRedfish / IPMIServers & BMCsMQTTIoT sensorsWired I/ODry contacts, RS-485

Node · in your facility

DSX ExchangeBMS/v1/PUB/#

Map & normalize

Check & publish

Facility equipment

Modbus

BACnet

SNMP

Redfish / IPMI

MQTT

Wired I/O

↓ Native protocols

Node in your facility

Map & normalizeCheck & publish

↓ DSX Exchange feed

DSX Exchange

What the Aravolta Node does for your site.

The node sits on your management network and talks to the Aravolta server, which runs fully air-gapped, on premises, or in the cloud, whichever your site requires. From there it does the work of a DSX Exchange publication in four stages.

The node arrives already knowing most of your equipment. More than 46,000 make, model, and firmware versions are in its library, so recognized devices start reporting without anyone writing a register map. Anything we have not seen before, we onboard within 48 hours at no cost to you. Custom gateways and programmed controllers use your approved point list, since the same controller model runs a different program at every site.

What your equipment becomes on the bus

Aravolta maps each device it finds to the objectType and pointType names the DSX Exchange BMS schema defines, so the software on the other side gets the same names and units from every site. A few of the most common cases:

Rack power, leak detection, and coolant temperatures go live first, since they are what the rest of the platform uses most. Facility-wide electrical and cooling context follows behind them.

What the published data looks like.

DSX Exchange speaks MQTT 3.1.1, and facility data follows the DSX Exchange BMS AsyncAPI schema. The schema is public on GitHub, and Aravolta checks every topic and payload against it before the feed goes live. Every point gets a live value topic and a retained metadata topic.

Here is one power meter point as Aravolta publishes it, value and metadata side by side:

1# Live value (QoS 0, publish-on-change)2BMS/v1/PUB/Value/PowerMeter/ActivePower/dal01/rpp-2a/kw-total3{ "value": 187.4, "timestamp": 1751457600000, "quality": 1 }4 5# Retained metadata6BMS/v1/PUB/Metadata/PowerMeter/ActivePower/dal01/rpp-2a/kw-total7{ "objectType": "PowerMeter", "pointType": "ActivePower",8  "objectName": "DAL01-RPP-2A", "objectId": "DAL01-RPP-2A",9  "servesId": ["DAL01-PDU-1"], "engUnit": "kW" }

Values publish on change at QoS 0 and republish every 100 seconds. Metadata is retained so new subscribers get context right away, and the publisher sends a heartbeat every 10 seconds. Topology is never guessed from tag names. Aravolta builds the objectId, servesId, and rackLocationId relationships from your single-line and rack IDs.

Publishing telemetry does not grant control. Write-back from DSX Exchange integrations is off by default. When you turn it on, your BMS stays the authority on every setpoint, as the DSX Exchange specification describes.

Four steps on your side.

Aravolta handles the discovery, mapping, and publishing. Your side is four steps.

From node install to DSX Exchange optimized

Every site is different, but the shape of a rollout is the same:

NVIDIA DSX Exchange is open, so you can watch your own feed from any subscriber on the network:

1mosquitto_sub -h <broker> -p 8883 \2  --cert client.pem --key client.key --cafile ca.pem \3  -t 'BMS/v1/PUB/#' -v

Commissioning is complete when the published model matches your site records and live field conditions.