Systems | Development | Analytics | API | Testing

A control plane for enterprise AI is the layer between the people and agents using AI and the infrastructure that runs it, meaning your models, GPUs, clusters, and cloud accounts. It decides who can use what, sends each request to the right resource, and records what that request cost. Klique is the control plane for enterprise AI. Teams get self-service access to approved tools, models, and GPUs, and IT sets the policies, quotas, and budgets behind them.

Most enterprise AI infrastructure conversations land on the same tension. Builders want self-service: instant access to GPUs, a stable endpoint for a model, a notebook, or a fine-tuning run that starts without a ticket. Platform teams want per-team quotas, identity-aware access, audit trails, tenant isolation, and cost attribution they can defend. Both sides are right, and most platforms make you choose between them because self-service and control get layered onto a base that was designed for neither.

The term “AI factory” is usually introduced as a hardware story: racks of accelerators, high-speed networking, and validated reference designs. That part is real, but it is not the part most organizations struggle with. The harder shift is operational. Moving from AI as a set of individual experiments to AI as a shared, governed capability that many teams depend on changes how compute is allocated, how environments are built, who owns the platform, and how the work is governed.

Two years ago, GenAI in production usually meant a single LLM serving a single endpoint. In 2026, it usually means much more. The applications shipping in front of users today are compound AI systems: orchestrated pipelines of retrievers, embedders, dialogue models, classifiers, code interpreters, SQL executors, and tools, with a single user request fanning out to several model calls across the stack.

AI sovereignty has become one of those phrases that sounds precise until someone asks what it actually means. For one federal agency, sovereignty means keeping sensitive data inside accredited boundaries. For another, it means running open-weight models in a FedRAMP-authorized private cloud. In defense and intelligence settings, it may mean operating inside an air-gapped environment at IL5 or IL6.

Almost every infrastructure team running modern AI is wrestling with the same question. The orchestrator their AI workloads want, Kubernetes, and the orchestrator their HPC environment was built on, Slurm, pull in different directions. This piece, drawn from the HPCKP 2026 session of the same name, looks at why that tension exists, what the workloads actually look like, how teams are bridging the two today, and where the pattern is heading.

Training models for physical robots is harder than training a typical model. The data has to be collected by hand through teleoperation, every change has to be tested on real hardware, and the loop from data to deployment runs constantly. In a recent proof of concept with a Singapore government agency, ClearML, Dell Technologies, and Hugging Face’s LeRobot framework turned that high-touch, manual process into an automated pipeline.

Most enterprises sized their AI infrastructure with a playbook written for training. However, training is no longer the typical workload. Inference now eats up roughly two-thirds of all AI compute, and it is changing shape fast enough that the rules of thumb from 18 months ago just do not hold. Our view at ClearML is pretty simple: when the workload shifts this much, the platform underneath it has to shift with it.

Running production LLM inference on a new accelerator family is a layered problem. The model matters. The runtime that exists for the GPU you have matters at least as much. So does the precision mode that works without losing accuracy, the inference engine that hits your throughput targets, and the secure endpoint the rest of your stack can actually call. The entire stack underneath the model is where most of the real engineering work lives and where the cost of getting it wrong shows up first.

NERSC, the mission high-performance computing center for the U.S. Department of Energy Office of Science, is using ClearML as part of the AI infrastructure stack for Perlmutter, the upcoming Doudna supercomputer, and the broader American Science Cloud. Here is a look at what they are exploring and why it matters for AI for science at scale.