Upbound Unfurls Control Plane for Managing AI Inference Workloads

Upbound, the primary maintainer of the widely adopted open-source Crossplane project, has launched Modelplane, an innovative open-source control plane tailored for artificial intelligence (AI) inference workloads. This new offering seeks to address the growing complexity and fragmentation within the infrastructure supporting AI models, particularly as they transition from experimental phases to core business operations. Modelplane extends the proven declarative management principles of Crossplane, which orchestrates cloud infrastructure, to the realm of AI inference. According to Upbound CEO Bassam Tabbara, IT teams can now leverage Modelplane to declaratively manage AI inference engines, integrating these operations into existing Kubernetes-based workflows. This means that defining a topology for an AI inference engine can lead to its automatic deployment on a Kubernetes cluster, similar to how other application workloads are managed. The tool is designed to simplify several critical aspects of AI inference management. It enables fleet-wide model scheduling, allowing organizations to determine where models run across available GPU capacity, various clouds, and different regions. Modelplane also supports autoscaling of model replicas based on demand and cost optimization targets, ensuring efficient resource utilization. Furthermore, it provides comprehensive infrastructure lifecycle management, covering the provisioning and management of clusters, nodes, and accelerators. Modelplane aims to streamline the integration of AI inference engines into cloud-native application environments, offering a unified gateway for routing inference requests and distributed caching for model weights. Available under an Apache 2 license, Modelplane is designed for deployment in both cloud and on-premises environments, offering flexibility and vendor neutrality. Upbound plans to further develop the project in the open with the community and intends to donate it to an open-source foundation later this year, reinforcing its commitment to a community-driven, vendor-neutral ecosystem for AI inference orchestration.