# How to Cut Your Cloud Bill: Tools by Savings Lever (2026) Canonical: https://www.pointfive.co/guides/how-to-cut-your-cloud-bill-tools-by-savings-lever-2026 The tools that cut cloud bills, by savings lever: commitments, idle resources, rightsizing, architecture, Kubernetes, data platforms, AI and allocation. By: PointFive Team Published: 2026-10-27 Updated: 2026-10-27 [All guides](https://www.pointfive.co/guides) # How to Cut Your Cloud Bill: Tools Ranked by Savings Lever (2026) PointFive Team October 27, 2026 12 min read ## TLDR - No single tool cuts a cloud bill. Savings come from eight levers, each with its own tools. - Start with the low-effort levers and your cloud provider's free recommenders. Larger, slower wins usually sit in architecture and configuration waste, which native tools mostly do not flag. - Pick by lever: commitments (PointFive, ProsperOps, Usage.ai, Zesty, CloudFix RightSpend), idle resources (PointFive, CloudFix), rightsizing (PointFive, Sedai), architecture and configuration waste (PointFive, CloudFix), Kubernetes (PointFive, Cast AI, ScaleOps, IBM Kubecost), data platforms (PointFive, Keebo, Unravel, Capital One Slingshot, Sedai), AI spend (PointFive OS, TokenShift by PointFive, LiteLLM, Portkey), and allocation (PointFive, CloudZero, Vantage, IBM Cloudability). - PointFive has two complementary products. PointFive OS, the agentic FinOps platform for cloud, data and AI, finds waste with 500+ deep detections across 40+ platforms, traces each finding to its technical cause and fixes it through IDE prompts or approved automation. TokenShift, the AI agent control plane for employee workstations, shows and caps what coding agents spend. - Order matters: remove idle resources before you commit, and set up allocation early so savings do not drift back. Where PointFive fits: PointFive is for teams that have worked through the native recommenders and still see the bill grow, because the remaining waste sits in architecture, configuration, Kubernetes, data and AI. PointFive appears in all eight levers below; for automated commitment purchasing, it sits alongside the specialists listed there. Nubank saved $8.2M in 9 months with PointFive. [See how Nubank executed 3,000+ DynamoDB remediations](https://www.pointfive.co/case-studies/nubank-dynamodb-optimization) . Disclosure: PointFive wrote this guide, and PointFive is listed first because it is our platform, not because of a ranking. Every competitor statement comes from that vendor's own website, documentation or press release, checked on October 27, 2026, and linked in the sources. Where a vendor publishes nothing, we say so. ## How we ranked the levers We order the levers by effort against typical impact. Effort is the engineering time and risk a change takes; impact is how much of a typical bill the lever touches. We quote savings percentages only when a cloud provider or vendor publishes one, and say who said it. | Rank | Lever | Typical effort | Typical impact | | --- | --- | --- | --- | | 1 | Commitments | Low | High on steady compute and databases | | 2 | Idle and orphaned resources | Low | Low to medium, but fast | | 3 | Rightsizing compute and databases | Medium | Medium | | 4 | Architecture and configuration waste | Medium to high | High, and mostly missed by native tools | | 5 | Kubernetes | Medium | High when clusters are a large share of spend | | 6 | Data platforms | Medium | High when Snowflake, Databricks or BigQuery is a top line item | | 7 | AI and LLM spend | Low to medium | Growing fast; depends on usage | | 8 | Allocation and accountability | High (organizational) | Indirect; keeps the other savings from drifting back | ## 1. Commitments: Savings Plans, Reserved Instances and CUDs What it is: You commit to a level of usage or spend for one or three years and pay a lower rate. It changes the price, not the usage. Typical effort: Low. No engineering changes. The risk is committing to capacity you are about to remove, so clean up idle resources first. What the native tools do: AWS says Savings Plans save up to 72% off On-Demand, and Database Savings Plans up to 35% on a one-year term. Microsoft says the Azure savings plan for compute saves up to 65% off pay-as-you-go prices. Google says resource-based committed use discounts save up to 55% on most machine series and up to 70% on memory-optimized ones. Each provider recommends purchases; you still decide, buy and manage the portfolio. Tools that act on it: - PointFive: consolidates Reserved Instances, Savings Plans and committed use discounts into one view across clouds, tracks effective savings rates, surfaces expiring commitments, and recommends coverage, so commitments are sized after the waste is removed. - ProsperOps: manages Savings Plans, RIs and CUDs autonomously across AWS, Azure and Google Cloud for "a small percentage of the realized savings"; Flexera acquired it in January 2026. - Usage.ai: buys and manages Savings Plans, RIs and CUDs on no-upfront terms across AWS, Azure and Google Cloud, offers cashback on eligible unused commitment value, and takes a share of new savings only. - Zesty: pairs AWS and Azure commitment optimization with Kubernetes optimization, on usage-based pricing. - CloudFix RightSpend: lowers the rate on EC2 you already run with no 1- or 3-year term; CloudFix says it gives 20-55% off EC2 and charges 25% of net-new savings. ## 2. Idle and orphaned resources What it is: Things you pay for that nobody uses: idle instances and databases, unattached disks, old snapshots, unused IP addresses, idle load balancers. Typical effort: Low per item; the work is confirming nobody needs it. Snapshot before deleting disks. What the native tools do: AWS Compute Optimizer gives idle recommendations for NAT Gateway, DynamoDB, ElastiCache, SageMaker endpoints and more, and its automation rules clean up unattached EBS volumes. Azure Advisor flags unattached disks, empty App Service plans and gateways with no connections. Google Recommender has idle VM, idle persistent disk, idle IP address and Cloud SQL idle instance recommenders. Each covers one cloud. Tools that act on it: - PointFive: PointFive OS detects idle EC2 and RDS, orphaned EBS volumes and snapshot bloat, unused Elastic IPs, idle load balancers, orphaned Azure disks, and idle Kubernetes deployments and orphaned PVCs across 40+ platforms, and fixes them through IDE prompts or approved automation. - CloudFix: stops idle EC2 instances and removes idle RDS, ElastiCache, EMR and SageMaker resources, idle NAT Gateways and unused Elastic IPs on AWS, running each approved fix as a Systems Manager Automation runbook. ## 3. Rightsizing compute and databases What it is: Matching instance size and type to real usage, including newer generations. Typical effort: Medium. Each change needs a performance check against peak load and a deploy window. What the native tools do: AWS Compute Optimizer rightsizes EC2, Auto Scaling groups, EBS, Lambda, ECS on Fargate, RDS and Aurora for free (enhanced metrics cost about $0.25 per resource per month). Azure Advisor recommends rightsizing or shutting down underutilized VMs and scale sets. Google's VM machine type recommender suggests resizing. All three recommend; your team implements. See [evaluating Compute Optimizer recommendations](https://www.pointfive.co/guides/ec2-right-sizing-compute-optimizer) . Tools that act on it: - PointFive: PointFive OS rightsizes EC2, RDS, Lambda memory, Azure VMs, Azure SQL (DTU and vCore) and Google managed instance groups at the template level, and fixes them under the approval level each customer sets. - Sedai: an "autonomous cloud platform" that tunes EC2, ECS, Fargate, Lambda, EKS, RDS and more, starting in Copilot mode, where you approve each action, before an optional switch to Autopilot. ## 4. Architecture and configuration waste What it is: Spend caused by how services are wired and configured, not how big they are: S3 traffic routed through a NAT Gateway instead of a VPC endpoint, cross-zone and internet data transfer, storage without lifecycle rules, verbose logs in expensive tiers, and services left in costly modes. Typical effort: Medium to high. Each finding needs investigation and usually an infrastructure-as-code change by the owning team. It is often the largest lever left once the basics are done. What the native tools do: Billing data shows NAT Gateway and data transfer line items, not which workload causes them. A few checks exist, such as Azure Advisor's Basic logs plan suggestion and Compute Optimizer's Aurora I/O-Optimized recommendation. The rest is left to investigation; see our [AWS bill investigation guide](https://www.pointfive.co/guides/why-your-aws-bill-jumped-2026-guide) . Tools that act on it: - PointFive: PointFive OS finds NAT Gateway waste through VPC Flow Log analysis, public internet access where VPC endpoints could be used, S3, Cloud Storage and Blob lifecycle and storage class gaps, Cloud Logging bucket configuration, and gp2-to-gp3 migrations, and traces each finding to its technical cause. - CloudFix: automates AWS-recommended configuration fixes such as EBS gp2-to-gp3 retyping, S3 Intelligent-Tiering and CloudWatch log retention, with nothing running until you approve it. ## 5. Kubernetes What it is: Pod requests set once and never revisited, poor bin packing across nodes, idle namespaces and GPU scheduling. Typical effort: Medium. Automation needs write access to the cluster, and pod changes affect reliability. What the native and open-source tools do: Karpenter (open source) provisions and consolidates nodes and picks spot or on-demand capacity, but never touches pod requests. OpenCost, a CNCF incubating project, measures and allocates cost without changing the cluster. See our [Kubernetes cost optimization guide](https://www.pointfive.co/guides/kubernetes-cost-optimization) . Tools that act on it: - PointFive: agentless, read-only discovery on EKS, AKS and GKE that finds workload sizing, node pool, idle deployment, orphaned PVC and DaemonSet overhead waste next to the cloud infrastructure underneath. - Cast AI: once automation is on, provisions and consolidates nodes, moves workloads to spot and sets pod requests; priced by quote, with free cost monitoring. - ScaleOps: rightsizes pod CPU and memory, sets replicas and HPA triggers, and consolidates nodes; self-hosted, priced by quote. - IBM Kubecost: cost allocation and visibility by namespace and workload, free up to 250 cores, with an optional controller that can rightsize Deployment requests (alpha). ## 6. Data platforms: Snowflake, Databricks and BigQuery What it is: Idle or oversized warehouses and clusters, inefficient queries, storage retention, and slots or bytes scanned. Typical effort: Medium. Warehouse and cluster settings change quickly; query and pipeline changes need data engineers. What the native tools do: Snowflake offers budgets, resource monitors, auto-suspend, Optimization insights and Adaptive Compute. Databricks recommends auto termination, compute policies, job compute for scheduled work and spot instances; its budgets mainly alert. BigQuery offers custom daily query quotas, maximum bytes billed and reservations with autoscaling. See the [best Snowflake cost optimization tools](https://www.pointfive.co/guides/best-snowflake-cost-optimization-tools-2026) . Tools that act on it: - PointFive: finds idle warehouses, auto-suspend gaps, oversized clusters, query inefficiency and Time Travel retention in Snowflake without accessing query text or data, all-purpose clusters that should be job clusters in Databricks, and BigQuery spend down to the table and job level. - Keebo: "Autonomous Optimization for Snowflake and Databricks" that changes warehouse size, auto-suspend and cluster count within guardrails you set, with a pay-as-you-go plan tied to credits saved. - Unravel: optimizes Databricks, Snowflake and BigQuery, testing query rewrites, compute resizes and configuration changes before they reach production and reverting them if behavior shifts; prices are not published. - Capital One Slingshot: Snowflake cost and performance management with warehouse rightsizing and query tuning; Capital One announced ML-driven Databricks job tuning in June 2025, but its current product page describes Snowflake only. - Sedai: rightsizes Databricks driver and executor nodes and termination timeouts in recommendation, approval or autonomous mode, without changing queries, schemas or code. ## 7. AI and LLM spend What it is: Model API calls, managed AI services such as Amazon Bedrock, Azure OpenAI and Gemini Enterprise Agent Platform, GPUs and inference endpoints, and the coding agents employees run. Typical effort: Low to medium for caching and batch; higher for model choice and architecture. What the native tools do: Anthropic and OpenAI both say their batch APIs cost 50% less than standard requests. Anthropic prices cache reads at 0.1 times the base input price for most models, and OpenAI says cached input tokens are discounted up to 95%, with prompt caching on by default for supported models. See the [best tools to track AI spend across teams](https://www.pointfive.co/guides/finops-tools-for-managing-ai-costs) . Tools that act on it: - PointFive OS: attributes AI cost by model, team, application and token type, and finds waste in Azure OpenAI provisioned throughput, idle Gemini Enterprise Agent Platform endpoints and notebooks, GPUs, inference endpoints, Databricks Model Serving and Snowflake Cortex AI, next to cloud and data spend. - TokenShift by PointFive: the AI agent control plane for employee workstations; it shows spend by developer, team, model and session for Claude Code, Codex, Cursor, GitHub Copilot CLI, OpenCode and Devin, sets model policies and spend caps by team, processes data on the device, and works alongside your gateway. - LiteLLM: an open-source gateway with budgets per key, user and team that fail requests once exceeded, plus Redis, semantic and storage-bucket response caches. - Portkey: an AI gateway, now part of Palo Alto Networks, with simple caching on all plans and cost- or token-based budget limits on Enterprise and select Pro plans. ## 8. Allocation and accountability What it is: Mapping every dollar to a team, product or customer so an owner sees it and acts. Typical effort: High, and mostly organizational: tagging, shared cost rules and named owners. Why it matters: Allocation cuts nothing by itself, but savings drift back when nobody owns the spend. What the native tools do: Each cloud groups cost by tags or labels and sends budget and anomaly alerts, such as AWS Budgets and Cost Anomaly Detection, one cloud at a time. Tools that act on it: - PointFive: tag-based cost allocation, Snowflake credit attribution and Databricks serverless cost attribution, with findings sent to Slack, Jira, Teams or ServiceNow and tracked through resolution. - CloudZero: allocates cloud, Snowflake and AI spend to teams, products and customers and calculates unit costs, sold as "a single subscription with all capabilities included". - Vantage: "the system of record for allocating and optimizing cloud, SaaS, and AI costs", with a free tier and public prices. - IBM Cloudability: enterprise allocation, budgets, forecasts and anomaly alerts, with Turbonomic-powered rightsizing actions in its Premium package. ## Side-by-side comparison | Lever | Typical effort | Native tools | Specialist tools | PointFive (PointFive OS + TokenShift) | | --- | --- | --- | --- | --- | | Commitments | Low | Recommend and sell plans | ProsperOps, Usage.ai, Zesty, CloudFix RightSpend | One view of RIs, Savings Plans and CUDs, expiry tracking and coverage recommendations | | Idle and orphaned | Low | Recommend; AWS automates EBS cleanup | CloudFix | Detects and fixes with approval | | Rightsizing | Medium | Recommend | Sedai | Detects and fixes with approval | | Architecture and configuration | Medium to high | A few checks | CloudFix (AWS fixes) | Core focus; traced to technical cause | | Kubernetes | Medium | Karpenter (nodes), OpenCost (visibility) | Cast AI, ScaleOps, IBM Kubecost | Agentless, alongside cloud | | Data platforms | Medium | Budgets, monitors, quotas | Keebo, Unravel, Capital One Slingshot, Sedai | Snowflake, Databricks, BigQuery | | AI and LLM | Low to medium | Batch, caching, provider limits | LiteLLM, Portkey | PointFive OS for AI services and GPUs; TokenShift for coding agents on workstations | | Allocation | High | Tags, budgets, anomaly alerts | CloudZero, Vantage, IBM Cloudability | Allocation tied to findings and owners | "Not listed" means we did not include PointFive for that lever in this guide. ## How to choose - You have never bought commitments: clear idle resources, then use native purchase recommendations or ProsperOps, Usage.ai or Zesty to automate them. - You have done the native recommendations and the bill still grows: the remaining waste is usually architecture and configuration. PointFive is built for this; CloudFix covers standard AWS fixes. - Most compute is Kubernetes: Karpenter for nodes plus Cast AI or ScaleOps for pods; OpenCost or IBM Kubecost for visibility; PointFive if Kubernetes is one part of a wider bill. - Snowflake, Databricks or BigQuery is a top line item: native controls first, then Keebo, Unravel, Capital One Slingshot or Sedai; PointFive when data spend sits alongside cloud and AI. - AI spend is growing: batch and caching first, a gateway for per-team budgets, PointFive OS for cloud AI services and GPUs, and TokenShift for coding agents on employee workstations. - Nobody owns spend: CloudZero, Vantage, IBM Cloudability or PointFive. See [cloud cost allocation tools](https://www.pointfive.co/guides/cloud-cost-allocation-tools) . Measure realized savings on the bill, not estimates in a dashboard. Deeper reads: the [best cloud cost optimization tools](https://www.pointfive.co/guides/best-cloud-cost-optimization-tools-2026) and the [best AWS cost optimization tools](https://www.pointfive.co/guides/best-aws-cost-optimization-tools-2026) . ## FAQ ### What tools can cut our cloud bill? Different tools work on different savings levers, so most companies combine a few: ProsperOps or Usage.ai for commitments, Cast AI or ScaleOps for Kubernetes, Keebo or Unravel for data platforms, and CloudZero or Vantage for allocation. PointFive covers all eight levers in one platform: PointFive OS tracks and recommends commitments, and finds and fixes idle, rightsizing, architecture and configuration, Kubernetes, data platform and AI waste across 40+ platforms, and TokenShift controls what AI coding agents spend on employee workstations. Start with your cloud provider's free recommenders before you buy anything. ### What is the fastest way to cut a cloud bill? Delete idle and orphaned resources, then buy commitments for the steady usage that remains. Both are low effort and use free native recommendations. ### Should I buy commitments or remove waste first? Remove waste first. A commitment locks in a rate for one or three years, so committing before cleanup means paying a discounted rate for capacity you were about to delete. ### Are native cloud tools enough? At modest, single-cloud spend, often yes. Beyond that, each covers one cloud, mostly recommends rather than acts, and rarely flags architecture and configuration waste. ### Do these tools change my resources automatically? Some do. CloudFix, Cast AI, ScaleOps, Keebo, Sedai and Unravel can apply changes within the permissions you grant; ProsperOps and Usage.ai buy commitments. PointFive discovery is read-only, and it uses write access only for changes the customer approves, under the approval level the customer chooses: answer, prepare a fix, or execute under per-change, per-playbook or no approval. ## Methodology and sources We reviewed each cloud provider's and vendor's own website, pricing pages, documentation and press releases on October 27, 2026, and did not use third-party sites for competitor claims. Facts reused from our other guides were checked against the same sources. If something is out of date, tell us at info@pointfive.co and we will correct it. - PointFive: [PointFive OS](https://www.pointfive.co/products/pointfive-os) , [DeepWaste](https://www.pointfive.co/products/pointfive-os/deepwaste) , [agentic remediation](https://www.pointfive.co/products/pointfive-os/agentic-remediation) , [TokenShift](https://www.pointfive.co/products/tokenshift) , coverage for [AWS](https://www.pointfive.co/coverage/aws) , [Azure](https://www.pointfive.co/coverage/azure) , [Google Cloud](https://www.pointfive.co/coverage/gcp) , [Kubernetes](https://www.pointfive.co/coverage/kubernetes) , [Snowflake](https://www.pointfive.co/coverage/snowflake) and [Databricks](https://www.pointfive.co/coverage/databricks) , [Nubank DynamoDB case study](https://www.pointfive.co/case-studies/nubank-dynamodb-optimization) , [AWS Marketplace listing](https://aws.amazon.com/marketplace/pp/prodview-xmaaw4f7qya3i) - AWS: [Savings Plans](https://aws.amazon.com/savingsplans/) , [Database Savings Plans](https://aws.amazon.com/about-aws/whats-new/2025/12/database-savings-plans-savings) , [Compute Optimizer supported resources](https://docs.aws.amazon.com/compute-optimizer/latest/ug/supported-resources.html) , [Compute Optimizer pricing](https://aws.amazon.com/compute-optimizer/pricing/) , [automation rules](https://aws.amazon.com/about-aws/whats-new/2025/11/aws-compute-optimizer-automation-rules) , [Budgets pricing](https://aws.amazon.com/aws-cost-management/aws-budgets/pricing/) , [Cost Anomaly Detection](https://aws.amazon.com/aws-cost-management/aws-cost-anomaly-detection/) - Microsoft: [Azure savings plan for compute](https://azure.microsoft.com/en-us/pricing/offers/savings-plan-compute/) , [Azure Advisor cost recommendations](https://learn.microsoft.com/en-us/azure/advisor/advisor-reference-cost-recommendations) - Google Cloud: [Compute Engine committed use discounts](https://docs.cloud.google.com/compute/docs/instances/committed-use-discounts-overview) , [recommenders list](https://docs.cloud.google.com/recommender/docs/recommenders) , [BigQuery cost controls](https://docs.cloud.google.com/bigquery/docs/controlling-costs) - Snowflake: [Managing cost](https://docs.snowflake.com/en/user-guide/cost-management-overview) , [Optimization insights](https://docs.snowflake.com/en/user-guide/cost-insights) , [Adaptive Compute GA](https://www.snowflake.com/en/blog/adaptive-compute-generally-available/) - Databricks: [cost optimization best practices](https://docs.databricks.com/aws/en/lakehouse-architecture/cost-optimization/best-practices) , [budgets](https://docs.databricks.com/aws/en/admin/account-settings/budgets) - Anthropic: [prompt caching](https://platform.claude.com/docs/en/build-with-claude/prompt-caching) , [batch processing](https://platform.claude.com/docs/en/build-with-claude/batch-processing) - OpenAI: [Batch API](https://developers.openai.com/api/docs/guides/batch) , [prompt caching](https://developers.openai.com/api/docs/guides/prompt-caching) - ProsperOps: [pricing](https://www.prosperops.com/pricing/) , [Flexera acquisition](https://www.prosperops.com/blog/the-next-chapter/) - Usage.ai: [homepage](https://www.usage.ai/) - Zesty: [homepage](https://zesty.co/) , [pricing](https://zesty.co/pricing/) - CloudFix: [homepage](https://www.cloudfix.com/) , [RightSpend](https://www.cloudfix.com/rightspend/) , [pricing](https://www.cloudfix.com/pricing/) - Sedai: [homepage](https://www.sedai.io/) , [data platforms](https://sedai.io/platform/data-platforms) , [pricing](https://www.sedai.io/pricing) - Kubernetes: [Karpenter](https://karpenter.sh/) , [OpenCost](https://opencost.io/) , [Cast AI](https://cast.ai/) , [Cast AI pricing](https://cast.ai/pricing/) , [ScaleOps](https://scaleops.com/) , [IBM Kubecost pricing](https://www.apptio.com/pricing?src=kc-com) - Keebo: [homepage](https://keebo.ai/) , [pricing](https://keebo.ai/pricing) - Unravel: [homepage](https://www.unraveldata.com/) , [pricing](https://www.unraveldata.com/pricing/) - Capital One Slingshot: [product page](https://capitalonesoftware.com/slingshot) , [Databricks announcement](https://capitalonesoftware.com/news/capital-one-software-strengthens-partnership-with-databricks) - LiteLLM: [budgets and rate limits](https://docs.litellm.ai/docs/proxy/users) , [caching](https://docs.litellm.ai/docs/proxy/caching) - Portkey: [caching](https://portkey.ai/docs/product/ai-gateway/cache-simple-and-semantic) , [budget limits](https://portkey.ai/docs/product/ai-gateway/virtual-keys/budget-limits) , [Palo Alto Networks acquisition](https://www.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-portkey-to-secure-ai-agents) - CloudZero: [Snowflake integration](https://www.cloudzero.com/integrations/snowflake/) , [pricing](https://www.cloudzero.com/pricing/) - Vantage: [homepage](https://www.vantage.sh/) , [pricing](https://www.vantage.sh/pricing) - IBM Cloudability: [product page](https://www.apptio.com/products/cloudability/) , [Advanced Rightsizing docs](https://www.ibm.com/docs/en/cloudability-commercial/cloudability-premium/saas?topic=premium-advanced-rightsizing) ## Related reading [All guides](https://www.pointfive.co/guides) - ### [Best Tools to Reduce OpenAI and LLM API Costs (2026): Gateways, Caching and Governance Compared](https://www.pointfive.co/guides/best-tools-to-reduce-llm-api-costs-2026) October 27, 2026 - ### [Best AI Agents for Cloud Cost Optimization (2026): Autonomy Levels Compared](https://www.pointfive.co/guides/best-ai-agents-for-cloud-cost-optimization-2026) October 9, 2026 - ### [Best Cloud Cost Optimization Tools (2026): An Honest, Comparison-Driven Guide](https://www.pointfive.co/guides/best-cloud-cost-optimization-tools-2026) June 2, 2026 --- Source: the public page above. Product screenshots and illustrative interfaces are examples, not live customer data.