A complete, factory-integrated NVIDIA GB300 NVL72 rack with 72 Blackwell Ultra GPUs and 36 Grace CPUs in a single NVLink domain, sold and supported by Cisco.
A complete, factory-integrated NVIDIA GB300 NVL72 rack with 72 Blackwell Ultra GPUs and 36 Grace CPUs in a single NVLink domain, sold and supported by Cisco.
This is rack-scale AI infrastructure ordered as one product. It is Supermicro’s NVIDIA GB300 NVL72 system, each rack contains: 18 1U compute trays, 9 NVLink switch trays, 8 power shelves, and in-rack management networking, integrated and tested as a complete rack before delivery. Cisco offers it under two part numbers that differ only in cooling scope: SRS-GB300-NVL72-M0 is the rack without any coolant distribution unit, for customers whose facility already provides one; SRS-GB300-NVL72-M1 includes in-rack CDUs and support in the offer. Cisco handles the selling motion, support through Cisco CX Services and Supermicro for liquid cooling. Management is done through Cisco Nexus One (Hyperfabric).
Each compute tray is packaged with two NVIDIA GB300 Grace Blackwell Ultra superchips, offering 72 Blackwell Ultra GPUs and 36 Grace CPUs connected through fifth-generation NVLink at 1.8 TB/s per GPU in a single rack. All 72 GPUs operate as one accelerator with roughly 21 TB of pooled HBM3e and about 38 TB of total unified memory including LPDDR5X. Target workloads are frontier-scale training and reasoning-model inference, where a single NVLink domain of this size replaces multi-node clusters. Liquid cooling is integrated into the platform; for information on any liquid cooling components, contact your Cisco sales specialist.
Table 1 summarizes the primary features and benefits of the platform.
| Feature | Benefit |
|---|---|
| 72x NVIDIA Blackwell Ultra GPUs in one NVLink domain | The whole rack behaves as a single accelerator for training and long-context inference |
| 36x NVIDIA Grace CPUs (72-core Arm Neoverse V2) | CPU and GPU share coherent memory over NVLink-C2C, removing PCIe bottlenecks |
| Approx. 38 TB unified memory (approx. 21 TB HBM3e) | Frontier-scale models and KV caches fit inside one rack |
| 9x NVLink switch trays at 1.8 TB/s per GPU | Non-blocking all-to-all GPU communication without external switching |
| 4x NVIDIA ConnectX-8 SuperNICs per tray at 800 Gb/s | Scale-out fabric bandwidth to grow beyond one rack when needed |
| NVIDIA BlueField-3 DPUs (up to 2 per tray) | Offloads north-south networking, storage, and security from the hosts |
| Factory-integrated, liquid-cooled rack | Arrives built and tested; direct-to-chip cooling handles the majority of the compute heat load |
| Two ordering options: without any CDU (SRS-GB300-NVL72-M0) or with in-rack CDU (SRS-GB300-NVL72-M1) | Match the offer to the facility instead of retrofitting the facility to the offer |
| Cisco CX Services support | Cisco support contract for a rack-scale, along with Supermicro support |
The rack is the computer: 72 GPUs delivered as one integrated system.
Most AI infrastructure is bought as servers and assembled into clusters on site. GB300 NVL72 inverts that: the rack is the unit of compute. Seventy-two Blackwell Ultra GPUs and thirty-six Grace CPUs arrive interconnected through nine NVLink switch trays at 1.8 TB/s per GPU, powered by eight 33 kW power shelves, and plumbed for direct-to-chip liquid cooling. What would otherwise be months of integration work is done before the truck arrives.
Cisco removes the remaining friction at the facility boundary. SRS-GB300-NVL72-M0 is for data centers that already operate a compatible coolant distribution unit and just need the rack. SRS-GB300-NVL72-M1 integrates an in-rack CDU and support into the same order. In both cases the liquid cooling scope is specified with your Cisco sales specialist, who can advise on in-rack, in-row, and liquid-to-air sidecar options and provide quotes for liquid cooling products. Cisco and Supermicro support contracts then covers hardware, and support.
Cisco offers this Supermicro platform in a defined rack configuration under two part numbers, listed in Table 2.
| Item | Selected configuration |
|---|---|
| PID, without CDU | SRS-GB300-NVL72-M0 (Supermicro-built GB300 NVL72 rack; facility provides the coolant distribution unit) |
| PID, with CDU | SRS-GB300-NVL72-M1 (Supermicro-built GB300 NVL72 rack with CDU support included in the offer) |
| Rack base | 1x NVIDIA GB300 NVL72 rack: 18x 1U compute trays, 9x NVLink switch trays, 8x 1U 33 kW power shelves |
| GPU/CPU | 72x NVIDIA Blackwell Ultra GPUs and 36x NVIDIA Grace CPUs (4x B300 and 2x Grace per tray) |
| Memory | Approx. 38 TB unified memory: approx. 21 TB HBM3e plus LPDDR5X |
| M.2 boot drives | 1x 1.9 TB NVMe M.2 per tray |
| NVMe storage | Qty (4) 3.84TB SED Drives, Up to 8x E1.S NVMe per tray |
| East-west networking | 4x NVIDIA ConnectX-8 SuperNICs at 800 Gb/s per tray |
| North-south networking | 1x NVIDIA BlueField-3 DPUs per tray |
| Management switching | (optional, purchased separately) 2x Cisco Nexus N93108TC-FX3 switches in rack |
| Liquid cooling options | In-rack CDU up to 250 kW, in-row CDU up to 1.8 MW, or liquid-to-air sidecar up to 200 kW; contact your Cisco sales specialist for liquid cooling component information and quotes |
| Support | Cisco CX Services and Supermicro Services |
| Availability | October 2026 |
Information about Cisco’s Environmental, Social and Governance (ESG) initiatives and performance is provided in Cisco’s CSR and sustainability reporting.
| Sustainability Topic | Reference | |
|---|---|---|
| General | Information on product-material-content laws and regulations | Materials |
| Information on electronic waste laws and regulations, including our products, batteries and packaging | WEEE Compliance | |
| Information on product takeback and resuse program | Cisco Takeback and Reuse Program | |
| Sustainability Inquiries | Contact: csr_inquiries@cisco.com | |
| Material | Product packaging weight and materials | Contact: environment@cisco.com |
Table 4 reflects the Supermicro data sheet for the NVIDIA GB300 NVL72 rack. Product specifications may change without notice.
| Category | Specification |
|---|---|
| Form factor | 48U rack-scale system; approx. 2236 x 600 x 1068 mm (88 x 23.6 x 42 in) |
| Compute trays | 18x 1U trays; each with 2x NVIDIA GB300 Grace Blackwell Ultra superchips (4x B300 GPUs, 2x Grace CPUs) |
| GPU | 72x NVIDIA Blackwell Ultra (B300) GPUs total |
| CPU | 36x NVIDIA Grace CPUs; 72-core Arm Neoverse V2 each |
| Memory | Per tray approx. 1.15 TB HBM3e plus 960 GB LPDDR5X; rack total approx. 21 TB HBM3e, approx. 38 TB unified |
| GPU interconnect | NVIDIA NVLink 5 via 9x NVLink switch trays; 1.8 TB/s per GPU, all 72 GPUs in one domain |
| Scale-out networking | 4x NVIDIA ConnectX-8 SuperNICs at 800 Gb/s per tray |
| In-band/out-of-band | Ethernet leaf switches in rack for in-band management; 1G/10G IPMI switch for out-of-band |
| North-south/storage | 1x NVIDIA BlueField-3 DPUs per tray |
| Storage | 2x 1.9 TB M.2 NVMe boot per tray; up to 8x E1.S PCIe 5.0 NVMe per tray |
| Cooling | Direct-to-chip liquid cooling across compute and switch trays; CDU options: in-rack up to 250 kW in-row up to 1.8 MW (multiple racks), or liquid-to-air sidecar up to 200 kW with no facility water required |
| Power | 8x 1U 33 kW power shelves (6x 5.5 kW PSUs each); operating power approx. 132 to 140 kW |
Table 5 lists what must be in place before installation. This is a rack-scale liquid-cooled system; for information on any liquid cooling components, or for a quote on liquid cooling products, contact your Cisco sales specialist.
| Requirement | Description |
|---|---|
| Facility space | Full-rack footprint approx. 2236 x 600 x 1068 mm; verify delivery path, floor loading, and service clearance for an integrated rack |
| Power | Facility feeds sized for approx. 132 to 140 kW operating power per rack, delivered through 8x 33 kW power shelves |
| Liquid cooling | SRS-GB300-NVL72-M0: facility-provided coolant distribution unit and water loop. SRS-GB300-NVL72-M1: in-rack CDU and support included; options scoped with your Cisco sales specialist |
| Scale-out fabric | Cisco N9100 Switches with NVIDIA Spectrum-X Ethernet infrastructure for multi-rack deployments |
The rack is a Supermicro-built system offered through Cisco under the two part numbers below. Certain liquid cooling components, such as in-row CDUs and liquid-to-air sidecards, are quoted separately: contact your Cisco sales specialist for information on any liquid cooling components or for a quote on liquid cooling products specifically. To order, contact your Cisco account representative or visit the Cisco Ordering Home Page.
| Part number | Product description |
|---|---|
| SRS-GB300-NVL72-M0 | Supermicro 48U liquid-cooled NVIDIA GB300 NVL72 rack, no In-Rack CDU |
| SRS-GB300-NVL72-M1 | Supermicro 48U liquid-cooled NVIDIA GB300 NVL72 rack with In-Rack CDU |
| LCS-SCDU-1K3LR001 | 1.8MW In-Row CDU, L2L, N+1 Pumps |
| LCM-SMLCPM3-3Y | Liquid Cooling Maintenance Service, SMLCPM3, 3-Year |
| LCS-SCDU-200AR001 | 200kW L2A Sidecar CDU, N+1 Pumps |
| SFT-ODLCRSUPSTD-5Y | BMC Software Support, Standard, 5Yr |
| SVC-NVSTDSUP-3Y | NVIDIA Enterprise Support Entitlement, 3Yr |
Warranty terms for this Cisco-offered Supermicro platform are standard Supermicro warranty terms: https://www.supermicro.com/en/support/warranty. Support is delivered through Cisco CX Services under the applicable service contract. Please contact your Cisco sales specialist for more details.
Cisco CX and Supermicro Services cover this platform from planning through deployment and ongoing operations, with dedicated support contracts. For more information, visit https://www.cisco.com/go/services.
Cisco Capital makes it easier to get the right technology to achieve your objectives, enable business transformation and help you stay competitive. We can help you reduce the total cost of ownership, conserve capital, and accelerate growth. In more than 100 countries, our flexible payment solutions can help you acquire hardware, software, services and complementary third-party equipment in easy, predictable payments. Learn more.
For more on Cisco AI infrastructure, visit https://www.cisco.com/go/ai. Full platform specifications are available on the Supermicro product page for the NVIDIA GB300 NVL72 system.
Learn more about Cisco and Supermicro partnership:
https://www.cisco.com/site/us/en/solutions/global-partners/supermicro/index.html