Voltage Park
GPU cloud provider offering on-demand NVIDIA H100, B200, GB300 clusters in Tier 3+ US data centers.
Updated April 2026
Overview
- Website
- voltagepark.com
- Segment
- GPU Cloud / AI Compute
Product overview
Voltage Park provides bare-metal access to NVIDIA HGX H100 (up to 36,000 GPUs), B200, and GB300 clusters with 3200 Gbps InfiniBand networking across 6-7 Tier 3+ US data centers, enabling fast AI training, inference, and HPC workloads with 99.982% uptime and robust security ]. Used by AI labs and startups like Cursor, Caltech, Phind, Imbue, and enterprises for scalable compute without long-term commitments ]. Distinct for owned hardware, affordable pricing starting at $1.99/hr on-demand, rapid self-serve deployment in 15 minutes, and mission-driven approach backed by $1B capital, recently merged with Lightning AI for integrated AI software ].
Revenue model
On-demand bare-metal H100: $1.99/hr Ethernet (1-1016 GPUs), $2.49/hr InfiniBand (8-1016 GPUs); long-term reserves (6+ months, 32-8000+ GPUs) contact for pricing; no hidden fees, pay-per-use to microsecond ], .
Moat
Voltage Park's key competitive moat is its ownership of a massive, high-performance GPU fleet—36,000+ NVIDIA H100, B200, and GB300 GPUs across 7 Tier 3+ US data centers with 3,200 Gbps InfiniBand networking—creating massive scale advantages and supply scarcity in AI compute that hyperscalers struggle to match at similar cost-efficiency. This is amplified by the 2026 merger with Lightning AI, integrating proprietary AI software (e.g., PyTorch Lightning, model serving, observability) with unlimited burst capacity, high switching costs from optimized bare-metal performance (40% faster LLM training), and network effects from 400,000 developers and $500M+ ARR customer lock-in.