Symphony as Compute Ontology: Extending Insight into OpenShift and NVIDIA AI Factories

Kubernetes and OpenShift have become the enterprise standard for platform compute, yet the container orchestration model provides no compute ontology, no typed, governed, semantically rich model of what the infrastructure is doing, how well it is doing it, or who governs what it does. NVIDIA's acquisition of Run:ai positions a GPU-aware container placer as the inference scheduling layer for its AI factory ecosystem but Run:ai operates at the same categorical level as Kubernetes, enhancing placement decisions without modeling the operational state of the workloads it places. The present paper, the first of two parts, proposes IBM Spectrum Symphony as the compute ontology for Kubernetes-based AI infrastructure and provides the architectural framework with comprehensive comparative analysis. The compute ontology provides typed resource metrics with semantic direction through ELIM, hierarchical consumer governance with sub-second rebalancing, SOAM service lifecycle management with per-phase failure policies, cross-substrate routing spanning OpenShift clusters, bare-metal GPU hosts, cloud burst instances and heterogeneous accelerators under one scheduling domain and GPU-accelerated data processing through Spectrum Conductor on the same EGO resource broker. A detailed comparison with Kueue (v0.17) demonstrates that Kubernetes' strongest governance extension operates at alpha API maturity with known defects, no periodic rebalancing and zero capability in workload perception, request routing, service lifecycle, cross-cluster federation, or data processing. A feature-complete analysis of Run:ai (v2.24) establishes that every Run:ai capability is replicable within Symphony's ELIM architecture, that Run:ai uses exclusively public GPU APIs (NVML, DCGM, CUDA) with no proprietary hardware access and that the NVIDIA AI Enterprise certification is a commercial license rather than a technical capability. NVIDIA's own selection of IBM Storage Scale (GPFS) at GTC 2026 for 10 petabytes of GPU-native analytics certified on DGX validates the coordination substrate the compute ontology uses. The architecture positions Symphony as the compute parallel to Palantir Foundry's data ontology, providing a vendor-independent compute ontology grounded in Songnian Zhou's 1987 doctoral research and four decades of production deployment exceeding hundreds of thousands of cores at individual financial institutions.