Frequently Asked Questions

MLCostIntel uses a read-only IAM role deployed via a CloudFormation template that you review and approve. The role has minimal permissions to read your Cost and Usage Reports (CUR) data. Setup takes under 5 minutes.

We never modify your infrastructure, and we use AWS STS temporary sessions — no long-lived credentials are stored in our system.

Setup takes under 5 minutes. You deploy a CloudFormation template to create a read-only IAM role, and MLCostIntel begins analyzing your Cost and Usage Report data immediately. Your first assessment is typically ready within minutes of connecting.

MLCostIntel analyzes all AI and ML-related AWS services including:

• Amazon SageMaker — training jobs, endpoints, notebooks, processing
• Amazon Bedrock — foundation model invocations and provisioned throughput
• EC2 GPU instances — p3, p4, p5, g4, g5 families
• EKS / Kubernetes — containerized ML workloads
• LLM API spend — OpenAI, Anthropic, and other providers

The Free tier includes connecting your AWS account, a full AI/ML spend breakdown, an optimization score with A–F grading, total savings identified by category, and the number of recommendations. It is free forever with no credit card required. See full pricing details.

The Monitoring tier includes everything in Free, plus full resource-level cost attribution, daily cost data refresh, cost anomaly alerts, savings roadmaps with implementation guides, and executive-ready assessment reports. It comes in three plans based on your ML spend:

• Starter ($500/mo) — up to $50K ML spend/mo
• Standard ($1,500/mo) — $50K–$150K ML spend/mo, plus priority support and team-level attribution
• Scale ($3,000/mo) — $150K–$500K ML spend/mo, plus Kubernetes attribution and multi-account support

The Enterprise tier includes everything in Monitoring, plus:

• Dedicated optimization support from our team
• Custom integrations and reporting
• SLA with guaranteed response times
• Quarterly business reviews
• Multi-cloud support

Enterprise is designed for teams spending $500K+ per month on AI/ML infrastructure. Contact us for custom pricing.

Yes. The Free tier is available forever with no credit card required. Connect your AWS account, get your optimization score, and see your AI/ML spend breakdown at no cost. Upgrade to Monitoring or Enterprise when the numbers make sense for your team.

Yes. MLCostIntel uses read-only access, encrypted data at rest and in transit, and tenant-isolated storage. We never store AWS credentials — we use temporary STS sessions. You control exactly what we can access through the CloudFormation template. Learn more about our security model.

No. MLCostIntel uses strictly read-only access. The IAM role we request has minimal permissions limited to reading your Cost and Usage Reports. We never create, modify, or delete any resources in your AWS account.

Yes. The Monitoring Scale plan and Enterprise tier include multi-account support, allowing you to analyze costs across your entire AWS organization. The Starter and Standard plans support a single AWS account. Contact us for details on multi-account configuration.

Generic FinOps tools group SageMaker and Bedrock with all other AWS services. MLCostIntel is purpose-built for AI/ML — it classifies every dollar by workload type (training, inference, development), tracks GPU utilization, attributes costs to specific experiments and models, and provides ML-specific optimization recommendations like spot instance strategies for training jobs and endpoint rightsizing.

Savings vary by organization, but common opportunities include:

• GPU rightsizing — 20–40% savings by matching instance types to actual workload requirements
• Spot instances for training — up to 90% savings on fault-tolerant training jobs
• SageMaker endpoint optimization — reduce over-provisioned inference capacity
• Unused resource cleanup — identify and eliminate idle GPU instances, forgotten notebooks, and orphaned endpoints
• Commitment discounts — Reserved Instance and Savings Plan strategies tailored to your ML workload patterns

Your free assessment will show your specific savings potential with a prioritized breakdown by category.

No. MLCostIntel is fully agentless. The only thing deployed to your account is a read-only IAM role via CloudFormation. There are no agents, daemons, or software to install, manage, or update.

CloudHealth is a general-purpose cloud cost management tool. MLCostIntel is purpose-built for AI/ML workloads with GPU utilization analysis, experiment cost attribution, LLM API tracking, and SageMaker-specific optimization that generic FinOps tools don't provide. See the full comparison →

Vantage is a multi-cloud cost observability platform for general infrastructure. MLCostIntel focuses specifically on AI/ML cost optimization — it classifies every dollar by ML workload type, tracks per-experiment costs, monitors GPU utilization, and provides ML-specific savings recommendations. See the full comparison →

CloudZero provides unit cost analytics and cost-per-feature insights for engineering teams. MLCostIntel is specifically built for AI/ML infrastructure — it classifies costs by ML workload type, tracks GPU utilization, attributes costs to individual training experiments, and monitors LLM API spend. If your cost challenge is specifically AI/ML infrastructure, MLCostIntel provides deeper, ML-native visibility. See the full comparison →

Kubecost focuses on Kubernetes cost allocation. MLCostIntel covers the full AI/ML cost stack including Kubernetes ML workloads, plus GPU instances, SageMaker, Bedrock, and LLM APIs. If your ML infrastructure extends beyond Kubernetes, MLCostIntel provides broader coverage. See the full comparison →

ML FinOps is the practice of managing and optimizing the financial operations of machine learning infrastructure. It extends traditional cloud FinOps with AI/ML-specific concerns like GPU utilization, training job efficiency, inference endpoint costs, and LLM API spend management. MLCostIntel is purpose-built for ML FinOps.