Hydra Host sells dedicated bare metal GPU servers for AI training, inference and HPC, and it does so without owning a single data center. Its control plane, Brokkr (the "AI Factory Operating System"), handles procurement, provisioning, billing, fleet monitoring and lifecycle management for independent data center operators, and a unified API lets customers rent that capacity anywhere in the network. The company describes itself as serving both sides of the market: infrastructure owners who want to turn power, facilities and hardware into revenue, and frontier labs, inference platforms, compute marketplaces and enterprises that want NVIDIA H100, H200, B200 and B300 systems with full hardware access. A Confidential Metal product, launched in beta in December 2025, provisions Intel TDX trust domains with GPU passthrough on bare metal in under an hour.
Scale has moved fast. During 2024 the network went to 3,706 nodes and 29,648 GPUs, revenue grew 18x and headcount went from 13 to 23. By June 2026 CEO Aaron Ginn wrote that Hydra runs across more than 60 data centers in 18 countries on five continents, has $2.8B in booked compute contracts, serves four of the five major inference platforms and keeps over 90% of customers. Named customers and partners include Parasail, Verizon Business, El Salvador's National AI Lab (a geothermal-powered NVIDIA HGX B300 deployment announced in November 2025), and hardware makers Lenovo, Dell, HPE and Supermicro. Postings describe the business as moving more than $200M in annual transaction volume.