Global – November 10, 2026 -- CIQ told delegates at AI Infrastructure Field Day that roughly 9,000 GPU nodes cost about $15.6 million in GPU hours for every 30 days they sat idle, according to a CIQ calculation based on one customer's deployment. The company used the figure to argue for building AI infrastructure from independent, replaceable layers instead of buying a single-vendor stack.
CIQ says AI stack layers change on one-to-three-month cycles
CIQ's presenter said every layer of the AI stack is changing at once, with some layers turning over every one to three months and hardware revised within a year. In the customer example, GPUs arrived before the rest of the data center was ready. Teams under production pressure often request a complete AI factory from one vendor, which commits them to that vendor's fixed view of the stack at one point in time.
CIQ's alternative, the AI kit, is the set of missing layers run next to infrastructure and tools a customer already operates. Each product must be useful without the others.
Rocky Linux from CIQ Pro AI deploys a GPU node in 3 minutes 44 seconds
Rocky Linux from CIQ Pro AI (RLC Pro AI) is the operating system layer. It combines Enterprise Linux with a current long-term kernel, tested NVIDIA drivers, CUDA and AI frameworks. CIQ puts happy-path node deployment at 3 minutes 44 seconds.
A delegate asked whether configuration management tools could deliver the same tuning. CIQ conceded they cover much of it but cannot change parameters set when the kernel and drivers are compiled. Someone must also repeat the work whenever a driver, CUDA or a framework releases. CIQ named teams that count GPU time as the buyers, including those with large GPU investments or those for whom a few percent more throughput changes results.
Fuzzball runs on top of existing Slurm and PBS deployments
Fuzzball, the workflow orchestration layer, uses an existing Slurm or PBS deployment as a resource manager, so nothing has to move on day one. It deploys on Kubernetes, so organizations not yet running Kubernetes must add it first. A single workflow definition runs on-premises and in the cloud. CIQ says this lets high-performance computing teams avoid building separate stacks for cluster, Kubernetes and generative AI workloads.
CIQ recommends starting with RLC Pro AI rather than adopting the full kit.