Best Databricks Cost Optimization Tools (2026): An Honest, Comparison-Driven Guide

TLDR

• On classic compute you pay twice: Databricks platform charges plus the cloud infrastructure underneath. Serverless bundles compute into the Databricks bill. Waste concentrates in idle all-purpose clusters, oversized job clusters, unowned serverless usage, Delta storage retention and model serving.
• Start with Databricks' own controls: system tables, budgets, compute policies, auto-termination, serverless usage policies and predictive optimization. 2026 added a Governance Hub cost page (Beta) and AI spend caps in Unity Gateway.
• Pick by gap: jobs and clusters tuned for you (Capital One Slingshot, Unravel, Sedai, SELECT, Keebo for SQL warehouses), spend shown, allocated and governed (LakeSentry, CloudZero, Vantage), and Databricks waste found and fixed as one part of a multi-platform cloud, data and AI bill (PointFive).
• The market moved: Sync Computing's team joined Capital One Software, DoiT acquired SELECT, which added Databricks in June 2026, and Flexera acquired Chaos Genius.
• PointFive is an agentic FinOps platform for companies whose Databricks bill is one part of a larger cloud, data and AI bill. PointFive OS finds Databricks waste, such as idle clusters and unused Delta storage, next to the cloud beneath it, while TokenShift sees and controls the AI agents employees run on their workstations.

Where PointFive fits: PointFive is for teams that already use Databricks' native controls and want waste found and fixed across Databricks and the cloud, data and AI platforms around it, with each change approved the way they choose. PointFive OS runs 500+ deep detections across 40+ platforms, and Nubank saved $8.2M in 9 months with PointFive. See how Nubank automated 3,000+ DynamoDB fixes.

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.

Databricks' own cost controls in brief

• Billing system tables: system.billing.usage holds account-wide usage, tied to jobs, warehouses and notebooks. system.billing.list_prices logs list prices, without negotiated discounts.
• Dashboards and Governance Hub: prebuilt usage dashboards, plus a Governance Hub cost page (Beta, August 26, 2026) with spend trends, cost drivers, tagged spend and read-only recommendations. Genie can answer cost questions.
• Budgets: by account, workspace or tag, with email alerts; "a monitoring mechanism, not a hard cap".
• Unity Gateway budgets: AI budgets (July 23, 2026) per user, workspace or account that stop further requests once exceeded.
• Serverless usage policies (Public Preview): formerly serverless budget policies. They tag serverless usage for attribution; they do not limit it.
• Compute policies: limit what compute users create, including a maximum DBUs per hour and required tags.
• Auto-termination and timeouts: classic compute stops after an idle period you set; serverless SQL warehouses default to 10 idle minutes, Pro and Classic to 45.
• Serverless jobs standard mode: the default; fewer DBUs on the same SKU, with about 4 to 6 minutes of startup latency.
• Photon: on by default for serverless and SQL warehouses, at a different DBU rate; compare cost per job.
• Predictive optimization: runs OPTIMIZE, VACUUM and ANALYZE on managed tables, billed as serverless jobs; default for accounts created from November 11, 2024.

What native tools leave to you: budgets do not stop compute, recommendations cover governance rather than tuning, fixes are manual, and the cloud bill under classic compute is out of view.

How we evaluated

We compared each option on six dimensions, unweighted:

1. Databricks depth: jobs, all-purpose clusters, SQL warehouses, serverless, storage, model serving.
2. Action: recommendations, or automatic changes with guardrails and approvals.
3. Access: read-only metadata, or write permissions on compute.
4. Breadth: Databricks only, other data platforms, or the wider cloud and AI bill.
5. Allocation: cost by team, job, user or tag.
6. Pricing model: as publicly stated.

The tools

1. PointFive (PointFive OS and TokenShift)

Category: Agentic FinOps platform (waste detection and remediation across cloud, data and AI)

Best for: Enterprises running Databricks alongside other clouds, data platforms and AI providers that want one place to find and fix waste.

PointFive OS runs 500+ deep detections across 40+ platforms, including AWS, Azure, GCP, OCI, Kubernetes, Snowflake and Databricks. On Databricks it flags all-purpose clusters that should be job clusters, idle time and autoscale settings, single-threaded workloads on multi-node clusters, unused tables and volumes, excessive Time Travel retention and oversized model serving endpoints. It attributes serverless cost to jobs, queries and notebooks, and checks the cloud beneath, such as S3 lifecycle policies against Delta Lake retention. Each finding is traced to its technical cause; teams investigate with Pointer, the AI agent in Chat, or the MCP server. TokenShift is the AI agent control plane for employee workstations.

• Strengths: Databricks and its cloud infrastructure in one view, next to Snowflake, cloud and AI spend. Setup is a notebook run once that creates a read-only service principal; query text analysis is opt-in. Fixes through IDE prompts or approved automation, at the approval level the customer chooses: per change, per playbook or none.
• Limitations: No self-serve tier or public price list. Automated remediation needs customer-authorized write permissions. If Databricks is your only large bill, a specialist below may be simpler.
• Pricing: Annual contract, custom quote. Listed on AWS Marketplace.
• Choose it if: Databricks is one of several large line items and the gap is fixing waste, not seeing it. Nubank saved $8.2M in 9 months with PointFive; Dolby is also a customer.

2. Databricks native controls (as of October 2026)

Category: Built-in cost visibility, attribution and controls

Best for: Every Databricks customer, as the baseline.

The controls listed above, with 2026 adding the Governance Hub cost page and Unity Gateway budgets.

• Strengths: First-party data, no procurement, real limits through compute policies and Unity Gateway budgets.
• Limitations: Budgets alert only; usage policies only tag; Governance Hub is Beta. Costs show at list price.
• Pricing: Included; predictive optimization bills on the serverless jobs SKU.
• Choose it if: Always, and first.

3. Capital One Slingshot, including Sync Computing (as of October 2026)

Category: Job optimization and cost visibility (tunes and shows)

Best for: Enterprises with many scheduled jobs that want per-job tuning with approval controls.

Slingshot for Databricks (announced June 2024) shows cost spikes and the "costliest users, jobs and workloads". In June 2025 Capital One Software it said "the Sync Computing team has joined Capital One Software" and that "Slingshot now offers intelligent, auto-pilot optimization for Databricks jobs", with customers choosing which jobs need approval. synccomputing.com now redirects to Slingshot.

• Strengths: Per-job automation with an approval option; Snowflake and Databricks side by side.
• Limitations: The Slingshot page now leads with Snowflake; supported Databricks clouds are not listed.
• Pricing: Not published.
• Choose it if: Job cluster tuning is your biggest leak and you want approval per job.

4. Unravel Data (as of October 2026)

Category: Observability and automated optimization (tunes and shows)

Best for: Teams that want code, job and cluster fixes across several data platforms.

Unravel's pitch is "Databricks, optimized at every layer. Automatically." Its FinOps Agent applies fixes such as cluster rightsizing, with autonomy set "from manual approval to full automation". Arvix AI, launched May 27, 2026 for Databricks, Snowflake and BigQuery, rewrites slow or expensive code, shows the diff and applies it in one click.

• Strengths: Code, configuration and clusters; AWS, Azure, Google Cloud and on-premises; SaaS, marketplace or your VPC.
• Limitations: Default automation scope is not spelled out. No price list.
• Pricing: Subscription or pay-as-you-go based on monthly DBU consumption; free health check (pricing).
• Choose it if: You want code-level fixes as well as cluster changes.

5. Sedai (as of October 2026)

Category: Autonomous cluster optimization (tunes)

Best for: Sedai users on AWS or Azure that want Databricks clusters tuned too.

Sedai launched Databricks optimization on June 10, 2025. Its solution sheet offers "Autonomous optimization for Databricks on AWS and Azure": cluster rightsizing, auto-termination, node count limits and policy compliance, executed with one-click approval behind policy-based guardrails.

• Strengths: Databricks tuning inside a broader cloud optimization platform.
• Limitations: No GCP. Sedai's Databricks capability page now returns an error.
• Pricing: Sedai Billing Units (1 per vCPU managed) on an annual commitment; Databricks rates not published (pricing).
• Choose it if: You want one vendor tuning cloud resources and Databricks clusters.

6. SELECT, now part of DoiT (as of October 2026)

Category: Cost observability with automated savings (tunes and shows)

Best for: Teams on Snowflake and Databricks that want allocation and automated savings in one tool.

DoiT acquired SELECT in January 2026; SELECT "will continue to be offered as an independent product". SELECT for Databricks reached general availability on June 10, 2026 on AWS, Azure and Google Cloud. It breaks spend down by workspace, job, cluster, user and query, and "automatically adjusts worker count and auto-scaling behavior" on SQL warehouses and all-purpose compute.

• Strengths: Allocation, budgets, idle insights and automated savings.
• Limitations: Automation does not cover job clusters. The changelog cites read-only access without explaining permissions for automated changes.
• Pricing: Custom quote, from $1,499/month (pricing).
• Choose it if: You want one specialist for Snowflake and Databricks.

7. Keebo (as of October 2026)

Category: Autonomous SQL warehouse optimization (tunes)

Best for: Teams whose Databricks spend is mostly SQL warehouses.

Keebo offers "Autonomous Optimization for Snowflake and Databricks". On Databricks it resizes and reconfigures SQL warehouses, needing CAN_MANAGE on each. Its docs say Databricks support "is currently in preview", on AWS, Azure and GCP.

• Strengths: Scoped permissions: no access to the system catalog, only views you create.
• Limitations: Preview; SQL warehouses only.
• Pricing: Pay-as-you-go from $0/month, where 1 DBU saved equals 1 Keebo Credit; custom Enterprise (pricing).
• Choose it if: SQL warehouse sizing is the problem and you want to pay from savings.

8. Revefi (as of October 2026)

Category: Observability and optimization for data and AI (tunes and shows)

Best for: Teams that need data quality and cost across several platforms.

Revefi's agents "autonomously manage cost, performance, and quality of your LLMs, agentic workloads and data warehouses", including Databricks, Snowflake, BigQuery and Redshift. A homepage demo applies its Databricks cluster recommendations (node type, Photon, single-node) to an ETL job.

• Strengths: Data quality and cost together; free Starter plan.
• Limitations: Databricks features are thinly documented; paid plans scale with spend.
• Pricing: Starter free under $50K combined annual spend; paid plans "% of data platform spend" (pricing).
• Choose it if: You also need data observability.

9. Flexera Data Cloud Optimization, formerly Chaos Genius (as of October 2026)

Category: Databricks and Snowflake optimization inside a FinOps suite

Best for: Flexera customers adding data platform optimization.

Flexera announced its acquisition of Chaos Genius on January 6, 2026, and lists Data Cloud Optimization as "Visibility and rightsizing for Databricks and Snowflake".

• Strengths: Data platforms inside a broad FinOps portfolio.
• Limitations: Flexera does not say whether it applies changes or recommends them.
• Pricing: Not published.
• Choose it if: Flexera is already your FinOps vendor.

10. LakeSentry (as of October 2026)

Category: Databricks cost observability with approved fixes (shows and governs)

Best for: Databricks-only teams that want published pricing.

LakeSentry is "a Databricks cost observability and performance optimization platform". It flags waste and anomalies with estimated savings, and teams approve or dismiss fixes with an audit trail. It runs as SaaS, a Databricks App or in your VPC, with read-only metadata access.

• Strengths: Public pricing, a free tier, and a Databricks App option where "nothing but a license check leaves your environment".
• Limitations: Supported clouds not listed; SOC 2 Type II "in progress"; no company details on the site.
• Pricing: Free (1 user); Standard $499/month; Pro $849/month, billed annually (pricing).
• Choose it if: You want a Databricks-only tool with a self-serve start.

11. CloudZero (as of October 2026)

Category: Cost allocation and unit economics (shows)

Best for: SaaS companies that need Databricks cost per customer or feature.

CloudZero lets you "Track Databricks spend by job, cluster, and team in near-real time", allocating usage even with incomplete tags, calculating unit costs and alerting owning engineers to anomalies.

• Strengths: Unit economics across cloud and Databricks.
• Limitations: Describes allocation and alerts, not optimization actions.
• Pricing: "Single subscription with all capabilities included"; custom quote (pricing).
• Choose it if: Finance needs Databricks unit costs next to cloud costs.

12. Vantage (as of October 2026)

Category: Cost visibility and allocation (shows)

Best for: Engineering teams that want self-serve Databricks reports next to cloud spend.

Vantage reads Databricks system tables through a small serverless SQL warehouse and a Data Reader service principal, and "does not have the ability to perform any write actions". Negotiated rates need system.billing.account_prices (Private Preview).

• Strengths: Public pricing with a free tier.
• Limitations: Reporting only; the SQL warehouse costs an estimated $84/month.
• Pricing: Starter free up to $2,500 tracked spend; Pro $30/month; Business $200/month; Enterprise custom (pricing).
• Choose it if: You want Databricks reports without a sales cycle.

Side-by-side comparison

"Not documented" means we did not find it on the vendor's own pages. "Model only" means the pricing model is public but not the rate.

How to choose

• Nothing tuned yet: import usage dashboards, enforce tags and DBU limits with compute policies, set auto-termination and assign serverless usage policies.
• Job clusters are oversized: Capital One Slingshot or Sedai; Unravel if code is also the problem.
• SQL warehouses and all-purpose compute dominate: SELECT or Keebo.
• Nobody owns the spend: LakeSentry, CloudZero or Vantage.
• Databricks is one of many large bills: PointFive, across Databricks, the cloud beneath it, Snowflake and AI. See how platforms price compute and our Snowflake tools guide.

Trial on real jobs and separate estimated from realized savings. See how to choose a cloud cost tool.

FAQ

What is the best tool to reduce Databricks costs?

Start with Databricks' own controls. For job and cluster tuning, consider Capital One Slingshot, Unravel or Sedai; for SQL warehouses, SELECT or Keebo; for allocation, LakeSentry, CloudZero or Vantage. If Databricks is one of several large bills, PointFive finds and fixes waste across it and 40+ other platforms.

Can Databricks budgets stop spending?

Only for AI traffic. Standard budgets alert and do not cap compute. Unity Gateway budgets stop further AI requests once exceeded. For compute, use compute policies and timeouts.

What happened to Sync Computing?

In June 2025 Capital One Software said the Sync Computing team had joined it to accelerate Slingshot's Databricks support. As of October 27, 2026, synccomputing.com redirects to Slingshot.

Does PointFive optimize Databricks?

Yes. PointFive detects idle and misconfigured clusters, single-threaded jobs on multi-node clusters, unused Delta storage and oversized model serving endpoints, attributes serverless cost, and reviews Databricks alongside the cloud beneath it.

Methodology and sources

We reviewed each vendor's own website, pricing pages, documentation and press releases on October 27, 2026, and did not use third-party sites for competitor claims. If something is out of date, tell us at info@pointfive.co and we will correct it.

• PointFive: Databricks coverage, PointFive OS, TokenShift, Nubank case study, AWS Marketplace listing
• Databricks: Unifying Databricks and cloud costs, Cost management tools, Billable usage table, Pricing table, Usage dashboards, Governance Hub cost page, Introducing Governance Hub, Budgets, Serverless usage policies, Unity Gateway AI spend controls, Compute policies, Compute configuration, Serverless notebooks, Serverless jobs, Photon, Predictive optimization, Budgets and alerts blog
• Capital One Slingshot: Slingshot, Databricks expansion, Databricks partnership and Sync Computing, Slingshot for Databricks docs
• Unravel Data: Databricks, pricing, news
• Sedai: Databricks solution sheet, launch press release, pricing
• SELECT: Databricks GA, SELECT joins DoiT, pricing
• Keebo: homepage, Databricks configuration, pricing
• Revefi: homepage, pricing
• Flexera: ProsperOps and Chaos Genius
• LakeSentry: homepage, pricing
• CloudZero: Databricks integration, pricing
• Vantage: Databricks integration docs, pricing