Reported by StorageReview SSD/NVMe/存储评测与行业新闻 on 2026-10-14. Summarized and analyzed by XAIPART Editorial.
What happened
Supermicro announced an expanded line of Open Compute Project (OCP) Accepted systems, racks, and rear door heat exchangers (RDHx), delivered through its Data Center Building Block Solutions (DCBBS) program and shown at booth B4 at the OCP Global Summit in San Jose through October 15.S1 The common element is the rack: the 21-inch ORv3 and the wider ORW both replace per-server power supplies and cabling with a centralized 48V DC busbar, which frees the space a conventional server spends on redundant power supplies for compute and cold plates.S1
The ORv3 rack series comes in 44OU and 48OU heights with integrated busbars, a certified static load of 5,000 lb (2,268 kg), and a liquid-cooling-ready mechanical design that now includes MGX 1.2 (Gen 3) racks.S1 Two propylene glycol coolants cover the technology cooling loop: DLC-1 for standard liquid-cooled racks and an ultra-low-conductivity DLC-2 fluid for specialty cold plates.S1 Both align with the OCP PG25 base specification so racks, in-rack CDUs, RDHx units, and cold plates run as a single cooling stack.S1
Residual heat is handled at the rack by a 4U in-rack coolant distribution unit (CDU) rated for up to 250 kW on 45°C warm water, a smaller sibling to the 1.8 MW in-row CDUs Supermicro ships with its Vera Rubin NVL72 racks.S1 Six rear door heat exchangers span 10 kW to 120 kW in AC and DC versions; DC variants are fed from the busbar, carry N+1 redundant fans, and include anti-condensation control.S1 Chiller doors cover the ORv3 rack at 20 kW to 100 kW and the ORW rack at 60 kW, managed over Redfish, SNMP, a web UI, and Modbus.S1 Supermicro has not published a density figure for a fully loaded ORv3 rack under that combination, so how far an older facility can go remains a site-by-site calculation.S1
On compute, the 2OU NVIDIA HGX B300 node places eight liquid-cooled GPUs in the 21-inch ORv3 envelope for up to 18 nodes and 144 GPUs per rack.S1 The 1OU FlexTwin packs two Intel Xeon 6 or AMD EPYC nodes into a single OU for HPC, and a 2OU four-node design reaches up to 45,696 Arm AGI CPUs per rack, targeted at always-on agentic AI.S1 The AMD Helios platform arrives as an ORW rack for inference, frontier-model training, fine-tuning, and sovereign AI.S1
The networking line sits on the same busbar and coolant loop. A 2OU co-packaged optics (CPO) switch built on Broadcom's Tomahawk 6 delivers 102.4 Tbps across 128 800G ports, accepts 40 to 60 VDC from the ORv3 busbar, connects to the rack loop through blind-mate QD08 couplings, and runs Supermicro Enterprise Advanced SONiC.S1 DCBBS also carries NVIDIA Spectrum-X Ethernet Photonics, a co-packaged Ethernet switch built on the 102.4 Tb/s Spectrum-6 ASIC with 200 Gb/s SerDes, which is in production now; NVIDIA cites 5x better power efficiency, 10x greater MTBI, 5x longer uptime for AI applications, and 1.3x faster time to deployment versus pluggable-transceiver designs, figures the source notes are vendor results that it did not audit.S1 SuperCloud Composer monitors and manages every DCBBS component, with hardware attestation and post-quantum cryptography algorithms listed for compliance with OCP security standards, and Supermicro says the OCP portfolio is available now.S1
| Metric | Reported value |
|---|---|
| ORv3 rack static load (certified) | 5,000 lb (2,268 kg) [S1] |
| ORv3 rack heights | 44OU and 48OU [S1] |
| In-rack CDU capacity | up to 250 kW on 45°C warm water [S1] |
| Rear door heat exchanger range | 10 kW to 120 kW (AC and DC versions) [S1] |
| HGX B300 GPUs per rack | up to 144 (18 nodes × 8 GPUs) [S1] |
| CPO switch aggregate bandwidth | 102.4 Tbps across 128 800G ports [S1] |
| CPO switch input voltage | 40 to 60 VDC [S1] |
| Spectrum-X vendor claims vs pluggable transceivers | 5x power, 10x MTBI, 5x uptime, 1.3x deployment speed (vendor-reported, not audited) [S1] |
| Density figure for fully loaded ORv3 under RDHx + DLC stack | Not disclosed [S1] |
Background timeline
- - Supermicro announces expanded OCP Accepted portfolio at OCP Global Summit in San Jose, displayed at booth B4 through October 15. [S1]
- - Source publication date for the announcement of the 250 kW in-rack CDU, the 102.4 Tbps CPO switch on Broadcom Tomahawk 6, and the ORv3/ORW 48V busbar stack. [S1]
XAIPART take
Technical impact
XAIPART analysis The 48V busbar and DLC-1/DLC-2 stack turn the ORv3 rack into a self-contained power and thermal island, but the practical constraint is the 4U 250 kW in-rack CDU: a single rack of 144 HGX B300 GPUs at typical AI training duty cycles will press that budget, so multi-rack deployments need to budget a CDU per rack or plan warm-water loop balancing accordingly. The 40 to 60 VDC input on the CPO switch and the blind-mate QD08 coolant couplings are the design details that matter most for serviceability, since hot-swap of a switch or a server should not require draining coolant.
Market impact
XAIPART analysis Bundling racks, CDU, RDHx, and CPO switching under a single DCBBS SKU list simplifies procurement but concentrates supply risk on one vendor; for buyers already standardized on ORv3, Supermicro's vertical integration is competitive pressure on Delta, Vertiv, and Rittal CDU/RDHx programs. The CPO switch on Broadcom Tomahawk 6 puts a 102.4 Tbps option into the ORv3 envelope at the same time Micas Networks released a 1RU Tomahawk 6 CPO box, which means buyers sourcing Tomahawk 6 silicon now have at least two form factors to qualify. Lead times for fully integrated ORv3 racks with CPO switches are not disclosed by the source and should be confirmed with Supermicro before committing to an AI-factory build-out calendar.
Action
XAIPART analysis For design engineers qualifying an AI rack today, request the DLC-1 versus DLC-2 compatibility matrix for the specific cold plates in use and confirm Redfish, SNMP, and Modbus coverage on the 250 kW CDU before integrating it into existing BMS workflows. Procurement should request a quoted lead time for a fully populated 48OU ORv3 with HGX B300 nodes and the Tomahawk 6 CPO switch, since the source did not disclose it; this is the gating item for any 2027 capacity plan. Site planners should treat the missing per-rack density figure as a custom engineering input rather than a catalog number.
XAIPART Forecast
XAIPART analysis XAIPART expects that if Supermicro publishes a quoted lead time for a 48OU ORv3 rack populated with HGX B300 nodes and a Tomahawk 6 CPO switch at or under 26 weeks by 2027-03-31, then OCP-accepted ORv3 will become the default reference design for new liquid-cooled AI sites in 2027; if the published lead time stays above 26 weeks at that date, traditional EIA racks with in-row cooling will retain the majority share of new AI deployments through 2027. XAIPART analysis
Facts are as reported by the cited source. The design-in, sourcing, and forward-looking statements above are XAIPART's analysis, not the source's claims. Uncertainty: lead times, pricing, and a per-ORv3 density figure under the combined RDHx and DLC stack are not disclosed; the Spectrum-X 5x/10x/5x/1.3x figures are vendor-reported and unaudited; single-source confirmation only.
Corrections: none at publication. This brief will be updated with a timestamp if the source corrects its report.
Alternatives et cycle de vie
| Référence | Relation | Cycle de vie | Fondement |
|---|---|---|---|
| — | Aucune alternative référencée DONNÉES REQUISES | — | Nécessite match_type avant que toute mention de prédécesseur ne puisse apparaître |
Les mentions « prédécesseur » et « alternative » proviennent uniquement d'un type de correspondance enregistré, d'un statut de cycle de vie ou du texte source. Lorsqu'aucune correspondance n'est enregistrée, aucune mention n'est affichée. Nos propres annonces ne constituent pas une référence de marché.
Références concernées
Aucune correspondance dans le catalogue. Cette information ne correspond à aucune référence actuellement listée sur XAIPART. Nous ne déduisons pas de correspondances et n'affichons pas de liste fictive.
Pression sur les coûts par dimension
| Dimension | FRED series ID | Unit | Lag | As of |
|---|---|---|---|---|
| Input cost | PCOPPUSDM, PALUMUSDM, PNICKUSDM | USD/mt | 96 d | |
| Supply/demand | PCU334413334413, IPG3344S | index(1982=100) / index(2017=100) | 46 d | |
| FX & energy | DEXCHUS, DCOILBRENTEU, DHHNGSP | rate / USD/bbl / USD/MMBtu | 6 d |
No free authoritative interface for: gold · silver · rare earth · polysilicon · lithium carbonate · market share — listed rather than approximated.
Comment ce contenu a été produit
Traçabilité du processus — récupération, contrôle de la source, modèle et contrôle
| Source fetched | StorageReview SSD/NVMe/存储评测与行业新闻 · www.storagereview.com · 10,950 chars · fetch status success |
|---|---|
| Source sanity (S1) | ok — passed before any model call (no s1_ block recorded) |
| Model / prompt | MiniMax-M3 · prompt Blog Agent - News Summary (v2.8-final) · 10,601 tokens · 31s · run completed |
| Gate | verdict pass · 6/6 applicable checks passed · spec v2.9 · C10 ratio 0.058213 · C10 rounds 0 |
| Time contract | 2 <time datetime> values in this brief — 0 violations (+08:00 or date-only) |
Chaque ligne ci-dessus est produite par le pipeline, et non rédigée à la main.
Commentaires
Soyez le premier à commenter.
🔐 Veuillez vous connecter pour publier un commentaire.
Vous n'avez pas de compte ? vous inscrire