CloudBolt at KubeCon 2026: Kubernetes Cost Optimization, AI Automation & Multi-Cloud Management

Key Concepts

• Cloud Management Platform (CMP): A unified software suite for provisioning, managing, and optimizing infrastructure across on-premises, public, and "NeoClouds."
• Kubernetes Right-Sizing: The process of automatically adjusting container resource requests and limits to optimize performance and cost.
• Vertical Pod Autoscaling (VPA): A mechanism to adjust the CPU and memory requests of pods based on actual usage.
• Horizontal Pod Autoscaling (HPA): A mechanism to scale the number of pods in a deployment based on observed metrics.
• In-place Pod Resizing: A Kubernetes feature allowing resource adjustments without restarting the container.
• FinOps: The practice of bringing financial accountability to the variable spend model of cloud computing.
• Model Context Protocol (MCP): A standard for connecting AI assistants to systems, data, and tools.
• Dogfooding: The practice of a company using its own products internally to test and improve them.

1. Company Overview and Portfolio

CloudBolt, a company with a 15-year history in cloud management, provides a comprehensive platform that abstracts the complexity of various infrastructure tools (e.g., Puppet, Ansible, Terraform). Following the acquisition of StormForge, the company integrated Kubernetes-specific optimization into its broader cloud management suite. The platform now serves as a "single pane of glass" for cost visibility, allocation, and automated resource management across diverse cloud environments.

2. Kubernetes Cost Allocation and Optimization

A major challenge addressed by CloudBolt is the "black box" nature of Kubernetes costs.

• The Problem: Organizations often struggle to attribute costs to specific teams or customers because Kubernetes clusters are frequently over-provisioned with "safety buffers" by developers, leading to single-digit utilization rates despite being "fully allocated."
• The Solution: By mapping cloud cost data to StormForge utilization metrics, CloudBolt provides container-level insights. This allows FinOps teams to identify exactly which customers or services are driving costs and where efficiency gains can be made.
• Key Statistic: Yasmin noted that some customers have realized double-digit millions in savings within the first three months of implementation by addressing low utilization.

3. Methodology: The HPA/VPA Algorithm

CloudBolt differentiates itself from native Kubernetes autoscalers (HPA/VPA) through a patented algorithm.

• The Conflict: Native Kubernetes documentation warns against using VPA and HPA simultaneously on the same metrics.
• The Innovation: CloudBolt’s machine learning-based approach adjusts pod requests while simultaneously updating the HPA target utilization. This preserves the desired scaling behavior while optimizing resource consumption.
• Validation: The company was recently awarded a patent for this specific HPA algorithm, developed by their internal team of PhDs.

4. Building Trust and Adoption

A recurring theme in the discussion is the "speed of trust." Because automated right-sizing can impact application stability, CloudBolt focuses on:

• Guardrails: Providing controls that allow teams to build confidence in the machine learning recommendations before moving to full automation.
• In-place Resizing: Leveraging the new Kubernetes GA feature for in-place pod resizing to avoid unnecessary restarts, which is critical for sensitive enterprise applications.
• Proactive Adaptation: The system can detect spikes and proactively adjust resources, moving beyond simple forecasting.

5. AI/ML Integration and Future Roadmap

• Conversational Interfaces: CloudBolt is integrating MCP support to allow users to interact with the entire portfolio via natural language. This enables users to ask questions like, "What is the top memory waste for this customer?" or perform provisioning tasks through chat.
• AI Workload Management: The platform is evolving to detect short-lived AI/ML jobs and map their usage patterns to provide accurate recommendations.
• GPU Right-Sizing: A key item on the upcoming roadmap is extending the current CPU/memory right-sizing capabilities to GPUs, addressing the high costs associated with AI training workloads.
• Standardization: The platform utilizes FOCUS (FinOps Open Cost & Usage Specification) to ensure that cost metrics are standardized across different cloud providers (AWS, Azure, GCP).

6. Synthesis and Conclusion

CloudBolt’s strategy centers on transforming infrastructure management from a manual, high-buffer process into an automated, data-driven operation. By combining deep machine learning for resource optimization with a unified management layer for diverse cloud environments, they enable organizations to achieve significant cost reductions without sacrificing system reliability. The shift toward conversational AI interfaces and GPU-specific optimization reflects the company's commitment to evolving alongside the modern, AI-heavy enterprise landscape.