FinOps / Cloud Cost Management 2026 — Vantage / Infracost / OpenCost / Kubecost / CloudZero Deep Dive

"Cost is not a non-functional requirement. In 2026 it is the most important functional requirement." — a FinOps lead

This article is a map of the FinOps market as of May 2026. Since 2023 GenAI workloads exploded and it became common for cloud bills to suddenly double or triple. As a result, FinOps shifted from "nice to have" to "without it the company goes bankrupt." The FOCUS (FinOps Open Cost and Usage Specification) 1.0 from the FinOps Foundation went GA in 2024, and from 2025 AWS / GCP / Azure all officially support FOCUS-formatted billing exports. Multi-cloud cost comparison finally became possible in a shared vocabulary.

This piece compares cost-visibility SaaS (Vantage, CloudZero, Cloudability), open-source visibility (OpenCost, Infracost, Komiser, Cloud Custodian), Kubernetes-only (Kubecost, Densify), automated optimization (Spot.io, ProsperOps, Granulate), anomaly detection (Anodot, Yotascale), and mid-market (FinOut). It ends with real cases from Toss / Kakao in Korea, Mercari / Cybozu in Japan, and a decision guide answering "which tool should our org pick?"

1. The 2026 FinOps Map — Why Cost Again

The word "FinOps" entered mainstream use in 2019 when the FinOps Foundation launched under the Linux Foundation. The core idea is simple. Cloud cost is a joint responsibility of engineering, finance, and the business, and we need shared language, shared data, and shared process to manage it.

Through 2023, FinOps was roughly "show the cost and reduce it." Then in 2024 GenAI workloads exploded. Startups commonly racked up OpenAI bills above several hundred thousand dollars a month. Companies running their own models paid four dollars per hour per H100 GPU, times eight GPUs, twenty-four hours a day — adding up to tens of thousands of dollars per month. CFOs started asking, every quarter, why the AWS bill went up thirty percent while revenue did not.

The 2026 FinOps market falls into five categories.

Multi-cloud visibility SaaS: Vantage, CloudZero, Cloudability (Apptio then IBM), FinOut
Open-source visibility: OpenCost (CNCF), Infracost, Komiser
Kubernetes-only: Kubecost (IBM), Densify
Automated optimization: Spot.io (NetApp), ProsperOps, Granulate (Intel)
Anomaly detection / forecasting: Anodot, Yotascale
Policy engines: Cloud Custodian (Capital One OSS)

Gartner forecasts that by 2026 eighty percent of the global top one hundred companies will have a dedicated FinOps team, and the FinOps Foundation itself crossed ten thousand member organizations.

2. FinOps Foundation + the FOCUS Standard

The FinOps Foundation was founded under the Linux Foundation in 2019. Initially a best-practices community, it became a de-facto standards body when it started the FOCUS (FinOps Open Cost and Usage Specification) in 2023.

The core idea behind FOCUS is simple. AWS exports billing data as CUR (Cost and Usage Report), GCP as Billing Export, Azure as Cost Management Export, Oracle as OCI Usage — and the column names and semantics are all different. The same concept of "resource id" is resource_id in AWS, resource.name in GCP, ResourceId in Azure. Multi-cloud companies had to normalize this each time.

FOCUS is the agreed vocabulary for these columns. 1.0 went GA in 2024 and major clouds began emitting FOCUS format natively in 2025. Key columns:

# Subset of FOCUS 1.0 core columns
BilledCost: actually billed amount (after discounts/credits)
EffectiveCost: per-unit cost after commitment discount allocation
ListCost: list price cost
ServiceName: normalized service name (e.g. Amazon EC2)
ServiceCategory: category (Compute, Storage, AI/ML, etc.)
ChargeCategory: Usage / Purchase / Tax / Adjustment / Credit
ChargeFrequency: One-time / Recurring / Usage-based
ResourceId: resource identifier
ResourceType: resource type
SkuId: SKU identifier
PricingUnit: pricing unit (GB-Mo, vCPU-Hour, etc.)

The meaning of FOCUS is not merely "unified format." It means less vendor lock-in. Switching from Vantage to CloudZero used to require rebuilding the data pipeline from scratch; now you move the FOCUS dataset as a whole. The number of companies loading FOCUS-formatted cost data into their own data warehouses (Snowflake / BigQuery) is growing fast. "Do cost analysis in SQL" is back in fashion.

3. Vantage — The Leader in Multi-Cloud Visibility

Vantage started in the US in 2020. Founded by former DigitalOcean alumni, it positioned itself from day one as "FinOps built by developers." After closing a Series B in 2024 it became the de-facto leader of multi-cloud visibility SaaS.

Strengths:

30+ integrations: AWS, GCP, Azure, Oracle, Alibaba, Snowflake, Databricks, Datadog, MongoDB Atlas, Confluent, Fastly, Linode, DigitalOcean. The differentiator is "not just clouds, but SaaS too, all in one place."
Cost Reports are intuitive. Without writing SQL, you build queries like "EC2 cost for the last 90 days, split by region and team tag" in the GUI.
Active Discounts: automatically finds unused RI/SP and misallocated commitments
Cost Anomaly Detection: both threshold-based and ML-based
Transparent pricing (a percentage of spend plus a minimum)

Weaknesses:

Automated remediation actions are limited. "Finds the issues but you need another tool or manual work to fix"
Enterprise RBAC / SSO are only on higher tiers
Kubernetes cost needs a separate integration (Vantage uses OpenCost under the hood)

Who should use it: 50-500 engineer orgs, multi-cloud, that also want SaaS spend in one place. Toss, Figma, Snap, PagerDuty are known users.

4. Infracost — Cost Diffs in IaC PRs

Infracost started in the UK in 2021 as open source. Where other FinOps tools show "cost already spent", Infracost is nearly the only tool that shows "cost about to be spent" at PR time.

How it works:

A developer opens a Terraform / Pulumi / CloudFormation PR
A GitHub Action runs Infracost
The bot leaves a PR comment with a diff like "this change adds $1,840/month"

# Basic Infracost usage
infracost breakdown --path=terraform/
# Example output:
# Project: my-infra/terraform
#
# Name                     Monthly Qty  Unit       Monthly Cost
#
# aws_instance.web_app
# ├─ Instance usage (m5.large)  730  hours       $70.08
# ├─ EBS volume (gp3)            30  GB-month     $2.40
# └─ EBS snapshots                5  GB-month     $0.25
#
# Project total                                  $72.73

CI integration (GitHub Actions):

# .github/workflows/infracost.yml
name: Infracost
on: [pull_request]
jobs:
  infracost:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: infracost/actions/setup@v3
        with:
          api-key: $INFRACOST_API_KEY
      - run: infracost breakdown --path=. --format=json --out-file=/tmp/base.json
      - uses: actions/checkout@v4
        with: { ref: $GITHUB_HEAD_REF }
      - run: |
          infracost diff --path=. \
            --compare-to=/tmp/base.json \
            --format=github-comment \
            --out-file=/tmp/comment.md
      - uses: infracost/actions/comment@v3
        with: { path: /tmp/comment.md }

The meaning of Infracost is "putting cost into the code-review loop." When a developer picks db.r5.4xlarge for an RDS instance and opens a PR, reviewers start asking "this adds $1,400 a month — do we really need that much?" The location of cost decisions moves from production-discovery time to PR time.

Infracost Cloud (SaaS) adds policy automation, an org dashboard, and guardrails. The OSS CLI is free; Cloud is priced per seat.

Who should use it: every team using Terraform or Pulumi seriously, especially platform teams that see dozens of IaC PRs per day.

5. OpenCost (CNCF) — The Kubernetes Cost Standard

OpenCost is the open-source project Kubecost donated to the CNCF in 2022. It entered CNCF Incubating in 2024 and is the de-facto standard for Kubernetes cost visibility.

How it works:

Run an OpenCost pod in the cluster
It scrapes Prometheus and collects per-node, per-pod, per-namespace CPU / memory / GPU / storage / network usage
Multiplies by cloud price lists (AWS / GCP / Azure) to derive cost
Exposes results as Prometheus metrics

# OpenCost install via Helm
helm repo add opencost https://opencost.github.io/opencost-helm-chart
helm install opencost opencost/opencost \
  --namespace opencost \
  --create-namespace \
  --set opencost.prometheus.internal.serviceName=prometheus-server

Key Prometheus metrics:

opencost_node_total_hourly_cost           # per-node hourly cost
opencost_load_balancer_cost               # LoadBalancer cost
opencost_pv_cost                          # PersistentVolume cost
opencost_namespace_total_cost             # namespace total
opencost_workload_cost_running            # per-workload cost

OpenCost stops at "show the cost." It does not reduce it. A common pattern is OpenCost + Grafana dashboards + Slack alerts for visibility, while reduction is delegated to other tools (Karpenter, KEDA, Spot.io).

Vendor adoption:

Vantage, CloudZero, Datadog Cloud Cost Management, Grafana Cloud all use OpenCost as a backend
So even if you do not install OpenCost yourself, using those products means OpenCost is running underneath

Who should use it: every team running Kubernetes. It is free and lock-in-free, so the rational default is to install it.

6. Kubecost — IBM Acquired (2024)

Kubecost is a Kubernetes cost SaaS started in the US in 2019. It is the parent company that built OpenCost and donated it to the CNCF. IBM acquired Kubecost in 2024 and it now ships as part of the IBM Cloud Paks portfolio.

OpenCost vs Kubecost:

OpenCost = free core, CNCF incubating
Kubecost = a commercial product layered on top with UI / dashboards / multi-cluster / anomaly detection / recommendations

Kubecost differentiators:

Multi-cluster aggregation: dozens of clusters in one view
Allocation views: split cost by namespace / label / deployment / team / product
Savings recommendations: right-sizing, idle-node detection, spot-conversion candidates
Budget alerts: per-team or per-namespace budget overrun alerts
Cluster Right-Sizing: node-pool composition advice

Impact of the IBM acquisition:

Tight integration with IBM Cloud / OpenShift accelerated
Some enterprise pricing changes (the Kubecost Free Tier is still around)
The roadmap tilted somewhat from "Kubernetes only" toward "OpenShift + IBM Cloud first"

Who should use it: 100+ node multi-cluster orgs, that treat namespaces as team or product boundaries, and want to chargeback.

7. CloudZero — Unit Cost Economics

CloudZero started in the US in 2018. While other FinOps tools focused on "reduce total cost", CloudZero put unit cost front and center.

What is unit cost? Examples:

"infra cost per customer per month" = $0.42
"cost per 1,000 API calls" = $0.018
"cost per analytical query" = $0.07
"cost per 1,000 AI inference tokens" = $0.0012

Total cost may not drop, but if unit cost drops, the business is getting healthier. The CloudZero pitch is that "the CFO wants unit cost, not totals, and unit cost should move with revenue."

Technically CloudZero:

Integrates AWS / GCP / Azure / Snowflake / Databricks / Kubernetes (via OpenCost)
Uses its own DSL called CostFormation to express arbitrary cost-allocation rules (how cost flows to products / customers / teams)
Separates AI / ML cost categories distinctly (strong on GenAI workload analysis)
Shows unit-cost trends side by side with revenue / usage metrics

Pricing is not public; generally scaled to annual cloud spend. Mid-market to enterprise focus, not seat-based.

Who should use it: SaaS companies where "ARR vs cloud cost ratio" shows up on a board slide. AI companies that need to prove "per-token inference cost is dropping every quarter."

8. Cloudability (Apptio/IBM) — Enterprise

Cloudability is a first-generation FinOps SaaS that started in 2011. Apptio acquired it in 2019, and when IBM acquired Apptio in 2023, Cloudability became part of IBM.

Position:

One of the oldest FinOps tools, dominant share in the enterprise
Sold bundled with IBM Apptio as an ITFM (IT Financial Management) suite
Enterprise-negotiated pricing, generally expensive

Capabilities:

Full AWS / Azure / GCP / Oracle / OpenShift coverage
Refined True Cost (list vs effective) analysis
ML-based rightsizing
Strong commitment management (RI / Savings Plans)
True Up: automatic alerts at commitment expiry

Weaknesses vs newer SaaS (Vantage, CloudZero):

Heavy UI / UX, steeper learning curve
Setup and configuration often need consulting
Slower pace of change (enterprise requirements come first)

Who should use it: $10M+ annual cloud spend, already using Apptio ITFM, with an existing IBM enterprise agreement.

9. Spot.io (NetApp) — Spot Instance Automation

Spot.io was previously named Spotinst. It started in Israel in 2015 and NetApp acquired it in 2020. The core pitch: "we manage the risk of spot instances, you take the 60-90% savings."

How it works:

Spot Elastigroup orchestrates AWS Spot, Azure Spot VMs, GCP Preemptible VMs
ML predicts price changes and reclamations
On a reclamation signal it pre-launches new instances and migrates workloads
Advertised at roughly "30% of on-demand price" on average

Product family:

Elastigroup: spot automation for general workloads (EC2 / VMSS)
Ocean: spot automation for Kubernetes node pools. Effectively a commercial competitor to Karpenter
Eco: commitment (RI / SP) management. Competes with ProsperOps
CloudCheckr: acquired in 2021, added FinOps visibility

2024-2025 trends:

AWS Karpenter grew quickly as open source, eroding some of Ocean's rationale
Still, Ocean remains valid for "orgs without Karpenter operators"
Re-branding under NetApp BlueXP internally

Who should use it: orgs that want to use spot but do not want to write the reclamation-handling code. Mid-market with no Karpenter operators.

10. Densify / ProsperOps / FinOut / Anodot / Yotascale — Others

Densify is a veteran out of Toronto, Canada since 2010. Its strength is Kubernetes / VM right-sizing recommendations. ML analyzes workload patterns and produces PRs like "this deployment only needs 0.3 CPU instead of 1 core, and 384 Mi memory instead of 512 Mi". Supports Terraform / Helm output.

ProsperOps started in the US in 2018. It focuses on a single thing: AWS RI / Savings Plans automation. It replaces human-driven commitment portfolio management with an algorithm, weekly buying and selling SPs to maximize "effective savings rate." Fees are a percentage of savings (typically 15-25%).

FinOut started in Tel Aviv in 2021 and established itself in the mid-market. Its strength is a dashboard builder that decomposes cost into business units. Often described as more reasonably priced than CloudZero, with a free tier for sub-10-person startups.

Anodot started in Israel in 2014. Originally a business-metric anomaly detection SaaS, it later applied the same ML engine to cloud cost. One of the most refined anomaly detectors in this space. It automatically explains "EC2 cost jumped 23% yesterday because 100 r5.large were added in us-east-1."

Yotascale started in the US in 2015. Strength is flexibility in cost allocation. An engineering manager can build "the cost the search team spent on ML inference over the last 30 days" in the GUI without writing SQL.

Tool selection matrix (summary):

Tool	Strength	Weakness	Pricing
Densify	k8s/VM right-sizing recs	weak visibility	enterprise
ProsperOps	AWS RI/SP auto-trading	AWS only	% of savings
FinOut	mid-market dashboards	weak automation	reasonable
Anodot	anomaly-detection ML	heavy UI	enterprise
Yotascale	flexible cost allocation	low brand awareness	seat + usage

11. Komiser / Cloud Custodian — OSS Inventory + Policy

Komiser is an open-source cloud resource inventory tool by Tailwarden (originally Mlabs). A SaaS version (Tailwarden) launched in 2024.

How it works:

Read-only access to AWS / GCP / Azure / DigitalOcean / Linode / OCI
Inventories every resource
Visualizes cost / tags / owners / anomalies on one screen
Surfaces insights like "47 EC2s without tags" or "12 EBS volumes stopped for over 30 days"

# Komiser CLI quick start
docker run -d -p 3000:3000 \
  -v $HOME/.aws:/root/.aws:ro \
  tailwarden/komiser:latest
# Open http://localhost:3000

Cloud Custodian is the policy engine Capital One open-sourced in 2016. CNCF incubating. It is "express cloud-governance rules in YAML, and run automated actions on violating resources."

# Cloud Custodian policy - auto-stop untagged EC2 after 30 days
policies:
  - name: ec2-untagged-stop-30d
    resource: aws.ec2
    filters:
      - State.Name: running
      - "tag:Owner": absent
      - type: instance-age
        days: 30
        op: gt
    actions:
      - type: notify
        to: [costops@example.com]
        template: untagged-instance.html
        subject: "Untagged EC2 - will stop in 7 days"
      - type: mark-for-op
        op: stop
        days: 7
      - type: stop

The value of Cloud Custodian is "manage policy as code." Rules like "per-region cost cap", "no untagged resources", "stop RDS idle for 7+ days" live in Git and change via PR.

Custodian users include Capital One itself, FINRA, HBO Max, eBay, and many other large enterprises.

Who should use it: Komiser for everyone (free, five-minute setup). Cloud Custodian for orgs with a security or governance team and a policy-as-code culture.

12. Granulate (Intel) — Workload Optimization

Granulate started in Israel in 2018 and is a workload-auto-optimization company. Intel acquired it in 2022 for $650M. Where other FinOps tools focus on "shrink the resource", Granulate took a different path: "run the same workload faster on the same infra, so cost drops as a side effect."

How it works:

Install a Granulate agent on the host
Analyze runtimes such as JVM / Python / Node.js / Spark / PostgreSQL / MySQL / NGINX
Auto-tune the OS scheduler, GC policy, and container cgroup parameters per workload
Advertises 20-40% CPU savings on average (large variance by workload)

Notable offerings:

gProfiler (open-source byproduct): continuous profiling on container hosts. Competes with Pyroscope / Polar Signals
Granulate Continuous Optimization: the commercial product
Optimizer for Spark: automatic tuning of Databricks / EMR Spark jobs
Optimizer for K8s: per-workload cgroup / scheduling tuning

After the Intel acquisition:

Re-branded under Intel Tiber Cloud Optimization
gProfiler stays open source
Spark / Databricks optimization cases grew (analytical workloads benefit most)

Who should use it: orgs running heavy runtimes like JVM / Spark / Python at scale. Past the right-sizing phase, looking for "what else can we squeeze."

13. showback vs chargeback / reserved capacity / savings plans

FinOps operating models fall into two main camps.

showback: only shows "your team spent $50K last month." The company still pays the bill centrally.

chargeback: the $50K is actually subtracted from the team's budget. The team carries P&L responsibility.

Most orgs start with showback. Starting chargeback without solid tagging / catalog hygiene triggers a flood of "that's not my cost" complaints. The usual path is 6-12 months of showback to build data trust, then a switch to chargeback.

Tagging standard (common pattern):

# Cost allocation tag standard example
team: payments              # responsible team
service: payment-api        # service name
env: production             # environment
cost-center: 12345          # accounting cost center
business-unit: consumer     # business unit
project: checkout-redesign  # project (optional)
owner: yj.kim@example.com   # contact

Reserved Instances (RI) and Savings Plans (SP) are AWS commitment discount programs. With a 1-year or 3-year commitment to a usage baseline, you get 30-72% off. RI is locked to instance family / region; SP (especially Compute SP) is far more flexible. The 2026 best practice is "cover 70-80% of baseline workload with 1-year Compute SP, rest on-demand or spot."

GCP's CUD (Committed Use Discounts) and Azure's Reserved VM Instances / Savings Plan have similar shapes. In multi-cloud each is managed separately.

14. Korea / Japan — Toss, Kakao, Mercari, Cybozu

Toss has operated a dedicated FinOps team since around 2022. It built an in-house cost dashboard and uses Vantage in some areas in parallel. According to Toss's published material, "Spark cluster right-sizing and RI optimization alone delivered savings of hundreds of millions of KRW per quarter." Korean fintech naturally has heavy data / AI workloads, and significant resources go into BigQuery / Snowflake cost management.

Kakao Group differs by entity — Kakao Enterprise, Kakao Bank, Kakao Pay each take different FinOps approaches. Kakao Bank loads cost data into its own data warehouse and analyzes via in-house SQL / BI. Kakao Enterprise manages its own cloud (KC) usage directly. Common pattern is heavy reliance on "open source + in-house build", with relatively low dependence on commercial SaaS compared to global peers.

Mercari is one of the most public Japanese companies for FinOps stories. The Mercari Engineering blog regularly publishes OpenCost adoption stories, GCP BigQuery cost reduction cases, and cost allocation in multi-tenant Kubernetes. Mercari runs an in-house cost dashboard (Mercari Cost Dashboard).

Cybozu is a Japanese SaaS company running Kintone / Garoon. It defines cost operations as "part of the production SRE team's role" and has publicly described using OpenCost + Datadog Cloud Cost Management. CyberAgent CIU (Cyber Infrastructure Unit) takes a similar approach with its own internal cost-visibility platform.

Common Korea / Japan patterns:

Commercial SaaS is expensive, so build-in-house is preferred
OpenCost + Infracost + loading into in-house BigQuery / Snowflake is the common combo
Commitments (RI / SP) are a joint decision area for finance and infra teams
Reaching chargeback is rare; most stop at showback

15. Who Should Pick What — Decision Guide

Recommended combinations vary by org size and workload.

Seed / Series A (1-20 engineers, less than $5K/month cloud spend)

AWS Cost Explorer + GCP Billing Console + Azure Cost Management is enough
Infracost OSS is free, install it
If costs suddenly spike, start with the Vantage free tier

Series B-C (20-100 engineers, $5K-$ 100K/month cloud spend)

Vantage or FinOut. Vantage if you want SaaS spend visible too
Lock Infracost into CI
Add OpenCost if you have Kubernetes
For spot, try Karpenter (OSS) first; fall back to Spot.io Ocean

Mid-market (100-500 engineers, $100K-$ 1M/month cloud spend)

Vantage or CloudZero (CloudZero if unit cost matters more)
Infracost Cloud for governance
OpenCost + Kubecost combo
ProsperOps for RI / SP automation
Consider Anodot for anomaly detection

Enterprise (500+ engineers, $1M+/month cloud spend)

Cloudability (Apptio) is the de-facto standard, or CloudZero / Vantage Enterprise
Cloud Custodian for policy automation
Densify for right-sizing automation
Granulate for workload optimization (large ROI on analytical workloads)
Load FOCUS-formatted data into your own Snowflake / BigQuery

Kubernetes-only orgs

OpenCost as the default
Add Kubecost when you need dashboards / multi-cluster
Right-sizing via Densify or your own VPA
Spot automation via Karpenter (OSS) or Spot.io Ocean

AI / ML companies

CloudZero is strongest at GenAI cost separation
GPU clusters: OpenCost plus in-house analysis
Track inference cost as a unit (per-token, per-request)
Consider Granulate for training-job optimization

The most important decision is not the tool itself but the political agreement on "who in our company gets to make which cost decisions." Without that agreement, any tool just adds more dashboards.

FinOps is not a tooling problem — it is an organizational problem. The difference in 2026 is that GenAI made cost no longer a "secondary engineering concern", and the FOCUS spec unified the vocabulary for multi-cloud cost analysis. The market is now arranged in five buckets. Regardless of the tool you pick, the start is simple. Install OpenCost, lock Infracost into CI, define a tagging standard, and run a quarter of showback data. Then decide whether to buy a SaaS or move to chargeback.

References

FinOps Foundation — https://www.finops.org
FOCUS Specification — https://focus.finops.org
Vantage — https://www.vantage.sh
Infracost — https://www.infracost.io
Infracost GitHub — https://github.com/infracost/infracost
OpenCost (CNCF) — https://www.opencost.io
OpenCost GitHub — https://github.com/opencost/opencost
Kubecost — https://www.kubecost.com
IBM Acquires Kubecost (2024) — https://www.ibm.com/blog/announcement/ibm-to-acquire-kubecost/
CloudZero — https://www.cloudzero.com
Cloudability (Apptio/IBM) — https://www.apptio.com/products/cloudability/
Spot.io (NetApp) — https://spot.io
Densify — https://www.densify.com
ProsperOps — https://www.prosperops.com
FinOut — https://www.finout.io
Anodot — https://www.anodot.com
Yotascale — https://www.yotascale.com
Komiser — https://www.komiser.io
Cloud Custodian (CNCF) — https://cloudcustodian.io
Granulate (Intel) — https://granulate.io
AWS Karpenter — https://karpenter.sh
Datadog Cloud Cost Management — https://www.datadoghq.com/product/cloud-cost-management/
State of FinOps 2025 — https://www.finops.org/insights/the-state-of-finops-2025/
Mercari Engineering Blog — https://engineering.mercari.com/en/blog/
Cybozu Tech Blog — https://blog.cybozu.io
CyberAgent CIU Blog — https://developers.cyberagent.co.jp/blog/archives/category/infrastructure/
Toss Tech Blog — https://toss.tech