Compute
Arise provides on-demand access to NVIDIA DGX B300 systems, offering either dedicated eight-GPU nodes with exclusive tenancy or flexible single-GPU rentals that scale with workload demands. Each system delivers 2.1 TB of HBM3e memory and 144 PFLOPS of FP4 sparse compute, with 800 Gb/s ConnectX-8 networking per GPU for cluster-scale orchestration. The service is currently available in the us-west-1 region.
Funding
Funding not disclosed
Founders
Founder details are not available yet.
Product
Problem
AI research and production teams increasingly require massive GPU clusters for training and fine-tuning large models, but procuring and managing dedicated high-end infrastructure is costly, time-consuming, and often results in underutilization or long lead times. Many cloud providers offer older-generation hardware or impose shared tenancy, complicating performance consistency and data security.
Solution
Arise delivers on-demand access to NVIDIA DGX B300 systems, making Blackwell Ultra-class compute available either as a dedicated node with eight GPUs on a single NVLink domain or as flexible single-GPU rentals that scale as workloads grow. Each system packs 2.1 TB of HBM3e memory and 144 PFLOPS of FP4 sparse compute, with every GPU connected at 1.8 TB/s over fifth-generation NVLink. The infrastructure uses ConnectX-8 SuperNICs delivering up to 800 Gb/s per GPU for cluster-scale orchestration, and air-cooled 10U packaging simplifies deployment in standard data centers. Customers get exclusive tenancy on dedicated nodes, ensuring no neighbors and no preemption, while the service is currently available in the us-west-1 region.
Target Audience
Primary customers are AI research labs, model developers, and enterprises that need high-throughput training and inference infrastructure for large language models, including workloads like Kimi K3, Nemotron 3 Ultra, DeepSeek V4 Pro, and GLM-5.3.
Features
- Dedicated DGX B300 nodes: eight NVIDIA Blackwell Ultra GPUs per system with exclusive tenancy and no preemption
- Flexible on-demand GPU rentals starting from a single Blackwell Ultra, scaling to full systems as needed
- 2.1 TB of HBM3e GPU memory per system with 62 TB/s total memory bandwidth
- Fifth-generation NVLink interconnect with 18 links per GPU and dual NVLink switches for full bi-sectional bandwidth
- 8× NVIDIA ConnectX-8 SuperNICs at up to 800 Gb/s each for InfiniBand or Ethernet cluster networking
- Dual BlueField-3 DPUs for storage and management traffic at up to 400 Gb/s
- Air-cooled 10U form factor with approximately 14 kW power draw, compatible with standard data center racks