LLM Inference

Kubernetes Wasn’t Built for GPUs. Make It Behave

Kubernetes counts whole GPUs and treats pods as disposable. An LLM pod is neither. Share the silicon with MIG/MPS/time-slicing and stop paying for idle ...

Sneha Gullapalli | August 12, 2026 | A100, AI infrastructure, AI Workloads, cloud native AI, Dynamic Resource Allocation, GPU autoscaling, GPU cost reduction, GPU optimization, GPU partitioning, GPU sharing, GPU time-slicing, GPU utilization, H100, Karpenter, KServe, Kubernetes DRA, Kubernetes GPU scheduling, LLM Inference, model caching, multi-instance GPU, NVIDIA GPU Operator, NVIDIA MIG, NVIDIA MPS, scale-to-zero, VRAM

Stop Treating GPUs Like Web Pods

Kubernetes schedules accelerators as opaque integers, and your bill pays for it. Share the silicon, scale on the right signal and keep weights out of the image ...

Veera Ravindra Divi | August 12, 2026 | AI infrastructure, AI serving, autoscaling, cloud costs, cloud native AI, DCGM exporter, DRA, Dynamic Resource Allocation, GPU costs, GPU scheduling, GPU sharing, GPU utilization, GPUs, inference workloads, KEDA, kubernetes, Kubernetes GPU scheduling, LLM Inference, MIG, model weights, MPS, NVIDIA GPUs, NVIDIA MIG, Prometheus, scale-to-zero, time-slicing

CNCF, cloud native, NVIDIA, AI, Peritus microservices

Deploying Docker AI Agents on OCI and OKE

This guide details the architectural transition of AI agents from experimental scripts to "first-class production workloads" using Oracle Cloud Infrastructure (OCI) and Oracle Kubernetes Engine (OKE). It emphasizes a zero-trust, scalable approach ...