RESOURCE PROPOSAL
01 / USB300
Silicon Valley
United States · California
192servers
1,536 GPUs
Gigabyte HGX B300
NEXFLOW MATRIX / COMPUTE
Compute for
ambitious workloads.
B300, GB300 and Vera Rubin options, with configuration and commercial coordination around your requirements.
GLOBAL RESOURCE NETWORK
Global reach.
Power at scale.
From Silicon Valley to Seoul. From Blackwell Ultra to Vera Rubin. Connect your AI ambitions with GPU infrastructure across key global markets.
04GLOBAL REGIONS
06GPU MODEL OPTIONS
GPUSERVERS & RACK-SCALE
RESOURCE LOCATIONS · GEOGRAPHIC REFERENCE
Explore the network.
RESOURCE PROPOSAL
02 / USB300
Atlanta
United States · Georgia
172servers
1,376 GPUs
Gigabyte HGX B300
RESOURCE PROPOSAL
03 / USGB300 NVL72
West Texas
United States · Texas
32racks
2,304 GPUs
Rack-scale NVLink
RESOURCE PROPOSAL
04 / CAB300
Toronto
Canada · Ontario
128servers / configuration
1,024 GPUs / configuration
Gigabyte / Dell configurations
RESOURCE PROPOSAL
05 / CAB300
Ontario
Canada
368servers
2,944 GPUs
Gigabyte HGX B300
RESOURCE PROPOSAL
06 / UKB300
United Kingdom
United Kingdom
256servers
2,048 GPUs
Supermicro B300
RESOURCE PROPOSAL
07 / ROVera Rubin NVL72
Romania
Romania
42racks
3,024 GPUs
NVIDIA Vera Rubin NVL72
RESOURCE PROPOSAL
08 / AEVera Rubin NVL72
Abu Dhabi
United Arab Emirates
28racks
2,016 GPUs
NVIDIA Vera Rubin NVL72
RESOURCE PROPOSAL
09 / KRB300
Seoul
South Korea
32servers
NVIDIA B300
RESOURCE PROPOSAL
10 / JPB300
Yokohama
Japan
32servers
NVIDIA B300
RESOURCE PROPOSAL
11 / IDB300
Jakarta
Indonesia
16–64servers
NVIDIA B300
RESOURCE PROPOSAL
12 / SGB300
Singapore
Singapore
2–64servers
NVIDIA B300
RESOURCE PROPOSAL
13 / MYB300
Malaysia
Malaysia
128servers
NVIDIA B300
Resource network includes capacity at different readiness stages. Locations and capacities describe sourcing options, not real-time stock or NEXFLOW-owned facilities. Toronto configurations are alternatives and are not added together.
Find the right location for your AI workload↗
JAPAN / KANTO · PROJECT BRIEFINFRASTRUCTURE IN FOCUS
Beyond the GPU.
A complete deployment plan.
Explore a Kanto-region B300 proposal combining dedicated GPU capacity with prefabricated liquid-cooled infrastructure.
128B300 SERVERS
1,024GPUs
4 MWPLANNED IT LOAD
Planning brief, September 2026. Image supplied as a modular pilot reference; not proof of completion of this project.
MULTI-GENERATION GPU PORTFOLIO
More architectures.
More possibilities.
From H100 and H200 to Blackwell and Vera Rubin. Explore GPU server and cluster resource options around your workload and preferred region.
01HOPPER
NVIDIA H100
GPU resource enquiries for training and inference workloads.
Discuss H100 resources ↗02HOPPER
NVIDIA H200
Explore resource options for memory-demanding AI workloads.
Discuss H200 resources ↗03BLACKWELL
NVIDIA B200
Connect with Blackwell-generation GPU server resource options.
Discuss B200 resources ↗04BLACKWELL ULTRA
NVIDIA B300
Explore GPU servers and dedicated cluster configurations.
Discuss B300 resources ↗05BLACKWELL ULTRA
NVIDIA GB300 NVL72
Explore rack-scale infrastructure and larger cluster resources.
Discuss GB300 NVL72 resources ↗06VERA RUBIN
NVIDIA Vera Rubin
Plan next-generation rack-scale AI infrastructure requirements.
Discuss Vera Rubin resources ↗GPU model portfolio. Regional availability and configurations are matched to each enquiry.
INFRASTRUCTURE / 部署形态
See the hardware.
Understand the setting.
GPU hardware and facility form are different layers. These reference photographs show a rack-room environment and a containerized enclosure.
01 / RACK ENVIRONMENT
Rack-room environment
Indoor racks with supporting power, networking and cooling infrastructure. The photograph illustrates a rack environment, not a specific NEXFLOW location.
02 / CONTAINERIZED
Containerized modular enclosure
A transportable enclosure brings infrastructure into a prefabricated module. Modular facilities can also use other building forms; not every modular site is a shipping container.
How this relates to our resource network
Toronto: the supplied brief describes an operating data centre with liquid-cooled servers.
West Texas: the supplied brief describes a modular liquid-cooled data centre; enclosure format is not specified.
Japan Kanto proposal: two prefabricated 2MW modules. Other locations: facility form is confirmed per resource.
Rack photograph: EPS. Modular image: supplied Japan project brief. Reference material; not a NEXFLOW facility ownership claim. View source ↗
CONFIGURATIONS / 配置与选型
Compare the building blocks.
Specifications below are drawn from the supplied project brief and require supplier reconfirmation.
NVIDIA B300 / Toronto
GPU per node8 × B300
ServerDell PowerEdge XE9780L / XE9785L
Host CPU2 × Intel Xeon 6767P; AMD option in brief
Memory3 TB DDR5-6400
Local storage4 × 7.68 TB NVMe
GPU network800G XDR InfiniBand
CoolingDirect liquid cooling
GB300 NVL72 / West Texas
GPU per rack72 GPUs
CPU per rack36 Grace CPUs
GPU memory in brief288 GB HBM3e / GPU
InterconnectNVLink 72-GPU domain
GPU network800G XDR InfiniBand
CoolingDirect liquid cooling
Rack power in brief140 kW
NEXT-GENERATION COMPUTE
Plan your next
architecture step.
Explore NVIDIA Vera Rubin alongside B300 and GB300 options. Connect with rack-scale resource options for next-generation AI workloads.
NVIDIA Vera Rubin NVL72
The NVIDIA Vera Rubin platform combines Vera CPUs, Rubin GPUs and NVLink interconnect in a rack-scale architecture for AI workloads.
Discuss your workload, target capacity, preferred location and project timeline with us.
Enquire about Vera Rubin ↗NVIDIA platform overview ↗Platform introduction only; this does not indicate NEXFLOW inventory, reserved capacity or a confirmed delivery date.
BEYOND THE GPU / 不止于 GPU
A cluster works
as a whole.
Bring the infrastructure requirements into the discussion early.
Connected at scale
Discuss GPU interconnect, front-end networking, storage networking and oversubscription against your distributed workload.
Designed for density
Review rack density, cooling design, facility power arrangements and acceptance conditions with the operator.
Ready for the data
Define dataset capacity, local scratch space, checkpoint throughput and data transfer requirements.
Let’s shape your solution.
Discuss your requirements ↗Hardware images: NVIDIA official product visuals. Platform reference images; supplied systems may differ. B300 visual shows DGX B300, while the Toronto proposal specifies Dell servers.