Nutanix is acquiring the French company Ryax Technologies. Ryax’s orchestration and scheduling technology is set to be integrated into future versions of the Nutanix Kubernetes Platform and Nutanix Enterprise AI, with the goal of making better use of GPUs and automatically placing AI workloads on the least expensive or most energy-efficient hardware.
Nutanix has announced the acquisition of Ryax Technologies, a French company focused on AI-driven compute orchestration. The technology is intended to help Nutanix achieve its ambition of running agentless AI everywhere: on-premises, in the public cloud, in neoclouds, and on HPC clusters.
The problem Nutanix is addressing is familiar. Companies purchase expensive accelerators, but these sit idle much of the time. According to Nutanix, this is because sizing, cluster selection, and reservations are often done manually and only once. Afterward, they are never reviewed again.
Right-sizing, fractional GPUs, and serverless
The Ryax capabilities that Nutanix aims to bring to NKP revolve around telemetry. The system learns historical resource profiles, adjusts GPU allocation per execution, and automatically handles out-of-memory errors by scaling up VRAM and restarting the job. In addition, there will be fractional GPU slices, allowing multiple containers to share the same card, plus serverless GPUs that retain capacity only during active compute cycles.
In Ryax’s own tests, per-execution sizing resulted in 62 percent fewer node-hours for a 30-run deep-learning burst, while the work was completed 5.7 percent faster. With Nvidia MIG-based fractional GPUs, four executions ran simultaneously on a single H100, reducing the cost per execution by 52 percent.
For NAI, Nutanix is planning a global meta-scheduling layer. This layer balances training, batch, and inference jobs against performance, cost, and energy goals and then routes them to Nvidia or AMD fleets, public clouds, or Slurm HPC clusters. Clusters are scored based on hardware power models, so that workloads can be directed to the cluster with the lowest expected energy footprint.
Tip: With NAI 2.8, Nutanix is focusing on AI agent governance