I/ONX offers a vendor‑neutral, high‑density AI compute platform that consolidates up to 64 accelerators into a single 42‑U node, eliminating most host CPUs and reducing the software stack to one OS instance. This design cuts rack‑scale total cost of ownership by 50% or more, lowers peak power consumption by up to 75%, and saves up to 75% of physical space, enabling enterprises to run production‑grade inference workloads more efficiently.
Funding
Funding not disclosed
Founders
Product
Problem
Enterprises deploying AI inference at scale face high capital and operational costs due to fragmented, host‑heavy architectures that require many CPUs, multiple OS instances, and large rack footprints. This “host tax” leads to excessive power consumption, low hardware utilization, and long ROI timelines.
Solution
I/ONX provides a vendor‑neutral, high‑density AI compute platform that collapses up to 64 accelerators into a single 42‑U node. By eliminating most host CPUs and consolidating the software stack into one OS instance, the Symphony SixtyFour platform cuts rack‑scale total cost of ownership by 50% or more and reduces peak power draw by up to 75%. The system uses unified DDR5 memory and open PCIe to support heterogeneous accelerators, delivering higher utilization and lower latency for production‑grade inference workloads. Simplified management and reduced infrastructure footprint enable faster deployment, lower OPEX, and a quicker return on investment for AI‑driven businesses.
Target Audience
Primary customers are large enterprises, cloud providers, and data‑center operators that run high‑throughput AI inference workloads and need to minimize infrastructure cost, power, and footprint.
Features
- Single‑node architecture supporting up to 64 AI accelerators (e.g., Furiosa, NVIDIA, others) via open PCIe
- Up to 94% reduction in host CPUs (from 16‑36 down to 2) and consolidation to one OS instance
- Unified DDR5 memory pool (up to 6 TB) providing a 2:1 RAM/VRAM ratio for NPUs
- Power-efficient design delivering ~13 kW peak for 64 accelerators versus >120 kW in traditional racks
- 75% space savings (one 42‑U node replaces a full 42‑48 U rack)
- Vendor‑neutral hardware stack reduces lock‑in and simplifies compliance licensing
- Integrated management tools that lower annual software licensing and OpenShift costs by >$500 k