Eleveight AI
Eleveight AI develops and operates AI Factories that provide sovereign, local access to high-performance NVIDIA GPU infrastructure for markets underserved by advanced AI compute. Its first facility in Gagarin, Armenia, runs on NVIDIA Blackwell B300 GPUs—the first deployment of this generation in the South Caucasus—and offers on-demand, dedicated, and inference services built to NVIDIA Reference Architecture standards. The data center achieves a Power Usage Effectiveness of effectively 1.1, placing it among the world's most efficient AI data centers.
- Artificial Intelligence
Funding
Founders
Founder details are not available yet.
Product
Problem
Organizations in emerging and underserved markets often lack access to advanced AI compute infrastructure, forcing them to rely on distant cloud providers with high latency, data sovereignty concerns, and limited local support. This gap hinders the development and deployment of large-scale AI models in regions where such capabilities are most needed.
Solution
Eleveight AI develops and operates AI Factories purpose-built for markets that need advanced AI infrastructure most. Its first AI Factory in Gagarin, Armenia, runs on NVIDIA Blackwell B300 GPUs, the first deployment of this generation in the South Caucasus. The company offers Neocloud on-demand GPU access, Dedicated reserved capacity, and Inference services, all built to NVIDIA Reference Architecture standards so workloads can be ported without custom configuration. The facility is designed for energy efficiency with a Power Usage Effectiveness of effectively 1.1, and scales from an initial 5 MW to a planned 40 MW capacity. Eleveight AI also provides CPU instances, storage options, and managed Kubernetes and Slurm services at no additional cost for on-demand users.
Target Audience
Primary customers are enterprises, governments, and research institutions in the Caucasus region and beyond that require sovereign, local access to high-performance AI compute for training, fine-tuning, and inference workloads.
Features
- NVIDIA B300 GPU instances with 288 GB vRAM, 18 PFLOPS at FP4, 8 TB/s bandwidth, and 1.8 TB/s NVLink for on-demand workloads at $9.45 per GPU-hour
- Dedicated bare-metal compute blocks for reserved capacity with performance and privacy guarantees
- Upcoming NVIDIA GB300 NVL72 and Vera Rubin NVL72 rack-scale systems for the largest training and agentic workloads
- NVIDIA Reference Architecture-compliant data center enabling workload portability without custom configuration
- Managed Kubernetes and Slurm on K8s included free for on-demand instances
- High-performance shared filesystem (WEKA) at $0.180 per GiB/month and object storage at $0.019 per GB/month
- Energy-efficient facility with PUE of effectively 1.1, among the most efficient AI data centers