CLOUDADV Optimizes Cloud VM Sizing with Zero-Shot AI Forecasting.

Cloud virtual machines are frequently overprovisioned, leading to unnecessary expenses and operational inefficiencies. A new interactive advisory system, CLOUDADV, aims to address this by assisting engineers in optimizing cloud instance sizing, even when workloads fluctuate. CLOUDADV leverages zero-shot time-series forecasting combined with bounded recommendation generation, offering planning horizons across days, weeks, and months. For each query, the system builds a comprehensive decision context, incorporating historical utilization, forecast summaries, current VM metadata, available instance options, pricing, and explicit sizing heuristics. The system uses a powerful LLM offline to generate reference recommendations, while a smaller production model is evaluated for deployment-time alignment, considering latency and cost. Case studies demonstrate substantial simulated monthly cost savings (over 50%) with minimal exceedance rates, indicating that zero-shot foundation models can effectively support decision-aligned provisioning in dynamic cloud environments, reducing the need for constant per-tenant retraining.