BLUF: NVIDIA's DSX (Data Smart Buildings Extension) partner roster, locked ahead of the June 1 Vera Rubin GTC Taipei keynote, has bifurcated the Big-4 HVAC OEM field into two named camps. Trane, Schneider, Siemens, Vertiv, and Procore are inside the standard. JCI, Honeywell, Carrier, and Daikin are outside. The difference is not capability — it is architectural posture. One camp ships composable layers that snap into NVIDIA's compute fabric. The other ships standalone product bundles that compete with it. For owners and operators evaluating an AI-HVAC commitment in 2026, the roster is the clearest piece of architecture evidence we have.
The roster, in one table
This is the partner list disclosed in DSX architecture briefings and confirmed in vendor partner-tier disclosures through late May 2026. Two notes: (1) "IN" means the vendor has shipped a DSX-aligned reference implementation or partner-tier integration; (2) "OUT" means the vendor has a competing standalone AI-HVAC product (Nantum, Forge, Niagara, Daikin Intelligent Equipment) but no public DSX integration as of this writing.
| Vendor | DSX Status | Architectural Posture | Public Evidence |
|---|---|---|---|
| Trane Technologies | IN | Composer — Cloud + ARIA + BrainBox stack on top of OEM substrate | 14,000+ commercial buildings, 25–40% savings band disclosed |
| Schneider Electric | IN | Composer — EcoStruxure + AVEVA digital-twin fabric, Aveva-Omniverse bridge | OpenUSD adoption on Aveva, GTC 2025 keynote co-presentation |
| Siemens | IN | Composer — Building X + Industrial Edge + Xcelerator marketplace | Xcelerator-NVIDIA Omniverse joint demos through 2025–2026 |
| Vertiv | IN | Composer — data-center thermal + power orchestration above multi-OEM hardware | NVIDIA reference-rack co-design, Phaidra×CoreWeave thermal proof points |
| Procore | IN | Composer (construction adjacency) — Agentic AI primitives May 2026 | ~3M users; Actions + Triggers shipped May 22, 2026 |
| Johnson Controls (JCI) | OUT | Productizer — Nantum AI standalone, Q3 2026 commercial rollout | Q2 FY26 +45% adj. EPS, +310bps EBIT margin (raised FY guide) |
| Honeywell | OUT | Productizer — Forge Building Performance suite, vertical-integrated | Forge revenue cited in FY24/25 earnings; no public DSX integration |
| Carrier | OUT | Productizer — Abound platform, retrofit reference 28% savings / $1.8M | Abound case studies; Tridium Niagara substrate remains closed |
| Daikin | OUT | Productizer — Daikin Intelligent Equipment, single-OEM stack | VRF lock-in via factory diagnostics; no public DSX integration |
Source mix: vendor partner-tier disclosures, GTC Taipei pre-keynote briefing materials, JCI Q2 FY26 earnings release (April 30, 2026), Trane Technologies investor day deck.
Why the roster matters more than the products
A partner roster is normally a marketing artifact. This one is not. DSX is an architectural standard that defines how a building's operating telemetry, control loops, and digital-twin state interoperate with NVIDIA's compute fabric — Omniverse for visualization and physics, OpenUSD for asset description, NIM microservices for model serving, and CUDA-X libraries for the optimization workloads. Being "IN" the standard means a vendor has agreed to expose interfaces at a layer that is above their own product. Being "OUT" means the vendor has chosen to keep the AI layer bundled with the OEM equipment.
Both choices are defensible. Composers (the IN camp) earn margin on the orchestration layer and concede that the substrate is increasingly commodity. Productizers (the OUT camp) defend margin by keeping the AI and the iron in one SKU. The architectural posture is the bet. For an owner with a multi-OEM portfolio — which is the typical Fortune 500 corporate occupier and the typical institutional REIT — composer-aligned tools fit; productizer-aligned tools force a single-vendor rationalization the owner did not ask for.
Three dated proof points the roster maps to
The bifurcation is not theoretical. It maps to three dated operational proof points landing in the same eight-week window:
- Trane Cloud + ARIA + BrainBox at 14,000+ commercial buildings — disclosed savings band 25–40%, IPMVP M&V Option C methodology referenced. This is composer-on-substrate proof at portfolio scale. The 14K figure was published in Trane's commercial backlog disclosure for Q1 2026.
- Phaidra × NVIDIA × CoreWeave × Applied Digital — 80% thermal-spike reduction at Max-Q — response time from minutes to under ten seconds. Data-center mode, not commercial real estate, but the architecture is the same: AI control above OEM thermal substrate. Phaidra is a Series B AI-HVAC pure-play; the integration runs on Vertiv hardware and NVIDIA inference.
- Delta Electronics, Omniverse-based functional twin at Taipei HQ — 20% energy savings disclosed. Same OpenUSD/Omniverse fabric, same architectural shape. Delta is also the 800 VDC AI-factory partner NVIDIA leans on; the Taipei HQ is the closest geographical and ownership analog to an APAC corporate occupier asking the AI-HVAC question this quarter.
Three altitudes — data center, corporate HQ, multi-building commercial portfolio — three dated proof points, one architectural pattern. The DSX roster is the procurement-facing artifact that names which vendors fit that pattern and which do not.
What the OUT camp is not wrong about
The fairer read on the OUT camp: vertical integration is a real moat in a market where 30% of commercial AI pilots fail to clear the demo-to-deployment line. JCI shipped a Q2 FY26 earnings beat (+45% adjusted EPS, +310bps EBIT margin, raised FY guide) because Nantum AI is a tightly-bundled product that the Field Services org can sell, install, and maintain without third-party integration risk. That is a real commercial schedule with real revenue, and it lands in Q3 2026 ahead of the next budget cycle. Honeywell's Forge revenue line is similar. Carrier's Abound has a reference case at 28% savings and $1.8M annual capture that owners can see and audit.
The risk for the OUT camp is not commercial execution. It is architectural drift. If the procurement-language layer (DSX, OpenUSD, NIM, Brick Schema, ASHRAE Guideline 36 + Addendum G) becomes how owners specify AI-HVAC by 2027, the productizers will face the same RFP question every quarter: "What is your DSX status?" And "we ship our own stack" stops being an answer when the question is about portfolio orchestration.
What a CRE buyer should do with the roster this week
Three actions, in order of practical leverage:
- Get a one-page architecture read of your own portfolio against the roster. For every building or asset, tag the existing HVAC OEM (Trane / JCI / Carrier / Daikin / Honeywell / Schneider / Siemens / mixed) and the existing BMS layer (Niagara / Honeywell EBI / Siemens Building X / open-Brick / mixed). The intersection tells you whether you are inadvertently locked to a productizer stack or whether you have composer optionality.
- Pull a written DSX-status answer from every shortlisted vendor in any AI-HVAC RFP issued between June 1, 2026 and the end of Q3 2026. "We are evaluating" is a no. "We are partner-tier with [named integration]" is a yes. Save the answers in the procurement file — they will matter in 18 months.
- Treat the JCI Q3 2026 commercial rollout as a budget anchor, not a vendor recommendation. JCI's Q3 launch is real, dated, and well-funded. Use it as the calendar reference for any 2026 AI-HVAC budgeting conversation — even if the eventual award goes to a composer-camp vendor. The procurement timeline is what is moving.
Where AISB sits on the roster
AISB's Fusion Model is composer-aligned by construction. The fleet sits above the OEM substrate: open-Brick for asset description, Speckle.dev and IFC 4.3 for coordination, ASHRAE-anchored claims, multi-CDE pattern detection across Procore / Autodesk Construction Cloud / Aconex / e-Builder / SharePoint fallback. A CRE owner running a multi-OEM portfolio uses AISB to orchestrate across the DSX-IN camp without inheriting any single composer's lock-in pattern. That is the explicit product posture, and it is the reason the roster matters to AISB and to anyone evaluating AISB against a single-vendor productizer.
Companion reading
- Orchestration Above OEM — the pillar page on the composer pattern
- AISB Proof — Trane / JCI / Phaidra / Delta operational evidence map
- Beyond Anthropic — owner-operator framing for AI in CRE
- Ask the agent: where does my portfolio sit on the DSX roster?
Methodology note: vendor partner-tier disclosures pulled from publicly released materials through 2026-05-29. Where a vendor's status is ambiguous, we have marked the column conservatively (Productizer) and welcome corrections via the agent at /ask/. Savings bands cited are vendor-disclosed; IPMVP M&V Option (A/B/C/D) referenced where the vendor has published a methodology note.