Request a quoteBuy GPUs
HYPERCUBE
HYPERQ
GPUs
Astra Cybercity
Sectors
Press Room
Company

info@worldoneindia.netWhatsApp +91 93244 08321

Buy NVIDIA GPUs.

Blackwell Ultra servers, NVL72 racks and desk-side stations from Supermicro, Gigabyte, HPE, Dell, MSI and ASUS. New, with manufacturer warranty, and priced per project.

A scalable unit of HGX Blackwell servers around a central network rack. Image: Supermicro.
15systems available to quote
6manufacturers, from Supermicro to ASUS
2 to 4 weekstypical HGX B300 lead time after KYC and escrow
Per projectpricing, issued against a named end customer
Blackwell Ultra

HGX B300 server nodes.

Eight Blackwell Ultra GPUs per node, joined by NVLink. Built by Supermicro, Gigabyte, HPE and Dell, air-cooled or liquid-cooled, with high volume available now.

Image: Supermicro.

High volume available

SupermicroNVIDIA HGX B300

SYS-822GS-NB3RT

8U air-cooled server with eight Blackwell Ultra GPUs and eight 800 Gb/s SuperNICs.

GPUs
8 × NVIDIA B300 SXM
GPU memory
288 GB HBM3e per GPU
CPU
2 × Intel Xeon 6 (e.g. 6768P, 64 cores)
System memory
DDR5-6400, 2 to 3 TB
Networking
8 × 800 Gb/s ConnectX-8 SuperNIC, BlueField-3 DPU
Cooling
Air
Form factor
8U
Lead time2 to 4 weeks after KYC clears and funds are in escrow

Image: Supermicro.

Strong; liquid-cooled allocation confirmed at order

SupermicroNVIDIA HGX B300

SYS-422GS-NB3RT-LCC

4U direct liquid-cooled server: the same eight GPUs in half the height.

GPUs
8 × NVIDIA B300 SXM6
GPU memory
288 GB HBM3e per GPU
CPU
2 × Intel Xeon 6788P (86 cores)
System memory
DDR5-6400, 2 TB
Networking
8 × 800 Gb/s ConnectX-8 SuperNIC, BlueField-3 DPU
Cooling
Direct liquid cooling
Form factor
4U
Lead timeConfirmed at order

Image: Gigabyte.

High volume available

GigabyteNVIDIA HGX B300

G894-SD3-AAX7

8U air-cooled server with the eight 800 Gb/s SuperNICs built onto the board.

GPUs
8 × NVIDIA B300 SXM
GPU memory
288 GB HBM3e per GPU
CPU
2 × Intel Xeon 6776P (64 cores)
System memory
DDR5-6400, 2 to 4 TB
Networking
Onboard ConnectX-8, 8 × 800 Gb/s OSFP; BlueField-3 DPU
Cooling
Air
Form factor
8U
Lead time2 to 4 weeks after KYC clears and funds are in escrow

Shown: Dell PowerEdge XE9780. Image: Dell Technologies.

HPE, Dell and other Tier-1 manufacturersNVIDIA HGX B300

Tier-1 HGX B300 nodes.

The same eight-GPU platform from HPE, Dell or another Tier-1 manufacturer, built to your specification.

HPE

HPE Compute XD690 S6G97A

Available

GPUs
8 × NVIDIA B300
GPU memory
288 GB HBM3e per GPU
CPU
2 × Intel Xeon 6776P (64 cores)
System memory
2 TB DDR5 (32 × 64 GB), up to 3 TB
Networking
8 × ConnectX-8 on the HGX board; BlueField-3 DPU; fabric quoted per design
Cooling
Air
Form factor
10U
Lead timeConfirmed at order

Dell or another Tier-1

Dell PowerEdge XE9780 or equivalent

On request

GPUs
8 × NVIDIA B300
GPU memory
HBM3e, per manufacturer
CPU
Intel Xeon 6 or AMD EPYC
System memory
Per buyer specification
Networking
ConnectX-8, BlueField-3 DPU
Cooling
Air or liquid
Form factor
Per manufacturer
Lead timeConfirmed at order
Blackwell and Hopper

HGX B200 and H200 nodes.

Eight-GPU nodes on the previous two NVIDIA generations, built by the Tier-1 manufacturer you prefer.

Shown: Supermicro SYS-A22GA-NBRT. Image: Supermicro.

Available; lower volume than B300

Tier-1 OEMNVIDIA HGX B200

HGX B200 node

Eight Blackwell GPUs in the Tier-1 system of your choice: Dell, HPE, Supermicro or Gigabyte.

GPUs
8 × NVIDIA B200 SXM, NVLink 5
GPU memory
180 GB HBM3e per GPU
CPU
2 × Intel Xeon 6 or AMD EPYC
System memory
DDR5, 2 TB typical
Networking
8 × 400 Gb/s NICs, BlueField-3 DPU
Cooling
Air
Form factor
Per manufacturer
Lead time2 to 4 weeks after KYC clears and funds are in escrow

Shown: Supermicro SYS-821GE-TNHR. Image: Supermicro.

Very limited; Hopper stock is scarce

Tier-1 OEMNVIDIA HGX H200

HGX H200 node

Eight Hopper GPUs for teams standardised on H200. Stock is scarce.

GPUs
8 × NVIDIA H200 SXM, NVLink
GPU memory
141 GB HBM3e per GPU
CPU
2 × Intel Xeon or AMD EPYC
System memory
Per buyer specification
Networking
8 × 400 Gb/s NICs
Cooling
Air
Form factor
Per manufacturer
Lead time4 to 6 weeks after KYC clears and funds are in escrow
Grace Blackwell

Rack-scale NVL72.

Seventy-two GPUs and thirty-six Grace CPUs in one liquid-cooled rack that works as a single NVLink domain. Delivered factory-integrated.

Shown: Supermicro SRS-GB300-NVL72. Image: Supermicro.

High volume available

Recommended

Tier-1 OEM, factory-integratedNVIDIA GB300 NVL72

GB300 NVL72

Seventy-two Blackwell Ultra GPUs and thirty-six Grace CPUs, liquid-cooled and joined in one NVLink domain.

GPUs
72 × NVIDIA Blackwell Ultra
GPU memory
288 GB HBM3e per GPU
CPU
36 × NVIDIA Grace (Arm)
System memory
LPDDR5X (Grace)
Networking
NVLink 5 rack domain; ConnectX-8 per compute tray
Cooling
Direct liquid, about 130 to 140 kW per rack
Form factor
Full rack
Lead timeConfirmed at allocation

Image: Supermicro.

Limited; we recommend GB300

SupermicroNVIDIA GB200 NVL72

SRS-GB200-NVL72-M1

Seventy-two Blackwell GPUs in eighteen 1U compute trays, with its own 250 kW coolant distribution unit.

GPUs
72 × NVIDIA B200
GPU memory
186 GB HBM3e per GPU (13.4 TB per rack)
CPU
36 × NVIDIA Grace (72 cores)
System memory
LPDDR5X (Grace)
Networking
NVLink 5; ConnectX-7 or BlueField-3 per compute tray
Cooling
Direct liquid; in-rack 250 kW CDU
Form factor
Full rack, 18 × 1U ARS-121GL-NBO trays
Lead timeConfirmed at allocation
Vera Rubin

Vera Rubin. The next generation.

NVIDIA’s next platform, with HBM4 memory, sixth-generation NVLink and 1.6 Tb/s ConnectX-9 networking. Allocation is limited, and first shipments are expected from December 2026 at the earliest.

A Vera Rubin NVL72 scalable unit of 1,152 GPUs. Image: Supermicro.

Shown: Supermicro SRS-VR-NVL72. Image: Supermicro.

Limited allocation

Tier-1 OEM, factory-integratedNVIDIA Vera Rubin NVL72

Vera Rubin NVL72

Seventy-two Rubin GPUs and thirty-six Vera CPUs in one liquid-cooled rack.

GPUs
72 × NVIDIA Rubin
GPU memory
288 GB HBM4 per GPU
CPU
36 × NVIDIA Vera
System memory
Per manufacturer
Networking
NVLink 6; ConnectX-9 SuperNIC, 1.6 Tb/s
Cooling
Direct liquid
Form factor
Full rack
Lead timeDecember 2026 at the earliest; January 2027 is more realistic

Shown: Supermicro 2U liquid-cooled HGX Rubin NVL8 system. Image: Supermicro.

Limited allocation

Manufacturer to be confirmedNVIDIA HGX Rubin NVL8

HGX Rubin NVL8 node

Eight Rubin GPUs on one NVLink 6 board, for halls built around eight-GPU nodes.

GPUs
8 × NVIDIA Rubin, NVLink 6
GPU memory
HBM4
CPU
Per manufacturer
System memory
Per manufacturer
Networking
Per manufacturer
Cooling
Air or liquid
Form factor
Per manufacturer
Lead timeDecember 2026 at the earliest; January 2027 is more realistic
RTX PRO Blackwell

Inference and enterprise servers.

PCIe GPU servers for inference, visual computing and simulation in a standard air-cooled data hall.

Image: Supermicro.

Available

Supermicro, or an equivalentNVIDIA RTX PRO 6000 Blackwell Server Edition

AS-5126GS-TNRT

5U air-cooled server with eight RTX PRO 6000 Blackwell Server Edition GPUs.

GPUs
8 × RTX PRO 6000 Blackwell Server Edition (PCIe 5.0)
GPU memory
96 GB GDDR7 per GPU
CPU
2 × AMD EPYC 9555 (64 cores)
System memory
768 GB DDR5-6400
Networking
ConnectX-6 Dx 100 GbE; more NICs optional
Cooling
Air
Form factor
5U
Lead time2 to 4 weeks after KYC clears and funds are in escrow
DGX Station architecture

A GB300 on your desk.

Desk-side AI systems built on NVIDIA DGX Station architecture: one GB300 Grace Blackwell Ultra Desktop Superchip with 748 GB of coherent memory. Offered on request while availability is confirmed.

Image: MSI.

On request; availability being confirmed

MSINVIDIA GB300 desk-side

XpertStation WS300

A desk-side AI supercomputer on NVIDIA DGX Station architecture.

Superchip
1 × NVIDIA GB300 Grace Blackwell Ultra
Memory
748 GB coherent: 252 GB HBM3e and 496 GB LPDDR5X
CPU
NVIDIA Grace
Networking
2 × 400 GbE ConnectX-8 SuperNIC
Power
1,600 W
Form factor
Desk-side tower
Lead timeTo be confirmed

Image: Supermicro.

On request; availability being confirmed

SupermicroNVIDIA GB300 desk-side

Super AI Station

ARS-511GD-NB-LCC

Liquid-cooled inside, in a tower that can also be rack-mounted as a 5U server.

Superchip
1 × NVIDIA GB300 Grace Blackwell Ultra
Memory
748 GB coherent: 252 GB HBM3e and 496 GB LPDDR5X
CPU
NVIDIA Grace, 72 cores
Networking
2 × 400 GbE ConnectX-8 SuperNIC
Cooling
Closed-loop direct liquid
Form factor
Tower or 5U rack
Lead timeTo be confirmed

Image: ASUS.

On request; availability being confirmed

ASUSNVIDIA GB300 desk-side

ExpertCenter Pro ET900N G3

ASUS’s DGX Station-class workstation, with room for an RTX PRO graphics card.

Superchip
1 × NVIDIA GB300 Grace Blackwell Ultra
Memory
748 GB coherent: 252 GB HBM3e and 496 GB LPDDR5X
CPU
NVIDIA Grace, 72 cores
Networking
2 × 400 Gb/s ConnectX-8 (QSFP112)
Power
1,600 W
Form factor
Desk-side tower
Lead timeTo be confirmed
Cluster networking and data-centre kit

Everything around the GPUs.

Switches, optics, cabling and cooling, supplied with the servers once the cluster design and site are confirmed. Quoted per cluster design, not as standalone items.

ItemModelDescriptionUsed for
InfiniBand switchNVIDIA Quantum-X800 Q3400144 ports of 800 Gb/s XDRGPU compute fabric
Ethernet switchNVIDIA Spectrum-X SN561064 ports of 800 GbEFront-end and storage network
Ethernet switchNVIDIA Spectrum-6 SN6600-LD128 ports of 800 Gb/s, liquid-cooledNext-generation fabric and spine
Management switchNVIDIA SN22011 GbE managementOut-of-band management
Optics800 Gb/s and 1.6 Tb/s OSFP transceiversCompatible modules; NVIDIA LinkX on requestGPU and switch links
CablingMPO-12 APC single-mode fibreCut to the lengths in the designFabric cabling
CoolingIn-row CDU, about 1.35 MW, N+1Serves liquid-cooled racksNVL72 and direct liquid-cooled halls
How to buy

Four details. Five steps.

Manufacturers release pricing and allocation only against a named project. These four details start a quotation.

  1. 1End customer

    The legal entity name and its company registration number.

  2. 2Installation site

    The data centre, delivery or colocation address and country.

  3. 3Indicative quantity

    The number of nodes or racks, and the models from this list.

  4. 4Target delivery date

    When the systems need to be on site.

Then, from agreement to order.

  1. 1Signed NCNDA

    A non-circumvention and non-disclosure agreement opens the file.

  2. 2Indicative quote

    Price and lead time for your models and quantity, for planning.

  3. 3KYC and end-use certificate

    The manufacturer’s checks on the end customer, site and use.

  4. 4Firm price and allocation

    Confirmed against the manufacturer’s allocation on the day KYC clears.

  5. 5Order and escrow

    You order direct with the manufacturer or supplier. Funds are held in escrow.

World One acts as introducing and facilitating broker. The supply contract is between the buyer and the manufacturer or supplier.

Configuration

The builds shown are the standard configurations we have priced. CPU, memory, storage and network options can be changed to your specification.

Availability

Availability and lead times are confirmed against the manufacturer’s allocation on the date KYC clears. Nothing is reserved by this list.

Delivery

Ex works, United States, unless agreed otherwise; delivered at place (DAP) to your site on request. Freight, duties, taxes, racks, power and installation are quoted separately.

Export control

Systems ship to the named end user and installation site only. No re-export or diversion.

Styling concept. Configuration not finalised.

No data hall? Take them in a module.

If you have power but no suitable building, the same GPUs can arrive inside a HYPERCUBE module: racks installed, liquid cooling connected and the system tested in the factory before it ships.

Questions.

Do you publish GPU prices?

No. Manufacturers release pricing and allocation only against a named project, so every quotation is written for a named end customer, site and quantity.

What do you need from me for a quotation?

The end customer’s legal name and company registration number, the installation site and country, the models and indicative quantity, and your target delivery date.

Who is the supply contract with?

With the manufacturer or supplier. World One introduces and facilitates the transaction, and funds are held in escrow.

Can the configuration be changed?

Yes. The builds shown are the standard configurations we have priced. CPU, memory, storage and networking can be changed to your specification before the firm price is issued.

Can the GPUs be installed for us?

Yes. Installation is quoted separately. You can also take the GPUs inside a HYPERCUBE module that arrives ready for power, cooling and network.

Why do export rules matter?

Advanced NVIDIA GPUs are subject to export regulations. The end user, the installation site and the end-use certificate decide what can be supplied and when, and systems may not be re-exported or diverted.

Ready to price your GPUs?

Tell us the end customer, the site, the models and quantity, and your delivery date.