Cloud Cost Tools in 2026: What They Attribute, and What They Can Change
Cost Explorer says EC2 went up. Every tool in this category exists to answer the next question, and they answer it at four different layers, against numbers that are not always what you paid.
Spend is up 38% this quarter and someone wants to know which feature did it. You open Cost Explorer. Cost Explorer says EC2 went up.
That gap is the entire category. CloudZero, Vantage, Infracost, nOps, OpenCost, Kubecost, CAST AI, ScaleOps: they all exist to answer the next question, and the pitches blur together because they all say “cloud cost.” They differ on three things that keep mattering after the pricing pages change.
What does the tool attribute against? Does it reconcile to what you actually paid, or to list price? And does it have write access to your infrastructure? Answer those three and the shortlist writes itself.
Vendor pricing and project status below were checked on 22 August 2026.
The four layers, and why Kubernetes is the hard one
Cloud billing stops at the node. You rent an instance, the scheduler crams pods onto it, and the bill knows about the instance while you care about the checkout service running at 12% CPU next to somebody’s batch job. Three problems fall out of that separation: splitting a shared node’s cost back to the workloads on it, the gap between what pods request and what they use, and the shared overhead nobody’s pod claimed.
That is why a Kubernetes-native tool is not the same product as a bill aggregator with a Kubernetes module, and why teams with serious cluster spend usually end up running two things.
Infracost: the only one your engineers will open unprompted
Infracost runs against a Terraform plan in CI and comments on the pull request: this change costs $847 more per month. That is the whole mental model, cost as pre-merge linting, and it is the easiest sell in the category because the number arrives where the decision is being made rather than in a dashboard nobody visits on a Tuesday.
The open source CLI is useful on its own and needs no account. It reads OpenTofu as well as Terraform, and there is Pulumi and CloudFormation support depending on how you wire it. AutoFix, added alongside their Series A, goes further and opens PRs that shrink what it flags.
What it cannot do follows directly from where it attaches. Once a resource is live, Infracost has nothing to say about who is using it, and anything created inside a cluster is invisible to it. Pricing has moved upmarket and off the public page: the OSS tier stays free, and anything serious is a conversation.
Vantage: the aggregator with prices you can actually read
Vantage normalizes AWS, GCP, Azure, Snowflake, Datadog, Databricks, MongoDB Atlas, Kubernetes and a long tail of SaaS bills into one queryable place. The integration count is the moat. So is the Cost Reports query language, which is SQL-shaped, version-controllable, and survives a reorg better than a hand-built dashboard. Network Flow Reports break AWS data transfer down by source and destination, which sounds dull until it finds the NAT gateway nobody could explain.
It is also the only tool here with published prices, which after a week of reading “contact us” pages counts for something. On the pricing page as of 22 August: Starter is free up to $2,500 of tracked spend with 3 users, Pro is $30 a month up to $7,500, Business is $200 a month up to $20,000, and Enterprise is custom with unlimited tracked spend. Tracked spend is what Vantage watches, not what you pay Vantage, and the flat monthly tiers are a meaningfully different shape from the percentage-of-spend model this category used to run on.
Where it stops is unit economics. You can slice by tag, environment, or service. Answering “what does it cost to serve this one customer” means writing custom queries and eventually hitting a wall.
CloudZero: the one built for the question finance asks
CloudZero’s whole product assumes the only cost question that matters is cost per business outcome. Cost per customer, per feature, per AI token. The allocation engine is configuration as code, which is how they get around the fact that nobody’s tagging is complete: you write rules that infer ownership from naming conventions, account boundaries, or call patterns instead of waiting for a tagging culture that is not coming.
Two things follow. The AI spend attribution is unusually coherent, because model API costs land in the same unit economics view as compute rather than in a separate spreadsheet. And the output is readable by people who control budget without a translation layer, which is half the value of any of this.
The cost of that is onboarding. It is consultative, it takes weeks before you see the view you bought it for, and at small scale you are paying for a framing you will not use. Pricing is quote-based, tiered against annual cloud spend, and CloudZero makes a point of it not being a straight percentage of your bill.
OpenCost: free, honest, and not reconciled
Most of the paid Kubernetes tools stand on this one. OpenCost is a CNCF incubating project that Kubecost created by donating its allocation engine, so running OpenCost means running the same math Kubecost sells. Point it at a cluster and within minutes you can answer what the payments namespace costs per day, broken down to the container.
Then read the sentence in the configuration docs that decides whether you can trust the number: “There is currently no reconciliation of the on-demand pricing with the billing data’s actual costs.” OpenCost allocates at list price. If you are deep into Savings Plans or Reserved Instances, every figure it shows is inflated relative to what you paid, and the ratio is not uniform across node types. You can override rates by hand in the Helm values, which works and is exactly as durable as any hand-maintained price table.
The rest of what it lacks is the usual open source shape: no long-term storage beyond the Prometheus retention you configure, no multi-cluster console, no RBAC or SSO. It shows. It does not act.
Kubecost, CAST AI, ScaleOps: reconciliation, then write access
Kubecost is the commercial wrapper: bill reconciliation against your real discounted rates, longer retention, SSO, RBAC, multi-cluster views, rightsizing recommendations. IBM acquired it and sells it under the Apptio line now, and the practical consequence is that pricing left the public web. The free tier is real and many teams never leave it, but it carries a core ceiling and a short retention window, and the IBM documentation blocks automated readers, so get the current numbers from a human before you architect around them. Third-party comparison posts quote figures that disagree with each other.
CAST AI, ScaleOps, and the autopilot half of nOps are a different risk category, because they run controllers inside your cluster that change requests, pack pods onto fewer nodes, and move eligible workloads onto spot. The savings are real and so is the fact that you have given an external system write access to production scheduling. CAST AI is the most visible and covers rightsizing, bin-packing, spot automation, and node lifecycle. ScaleOps is narrower and sharper: continuous pod rightsizing against live usage, which is the right shape if your actual problem is that nobody has maintained a request value in two years.
Neither publishes pricing any more. CAST AI’s pricing page as of 22 August is a form. Older comparisons quote a per-CPU rate and a monthly floor; I could not confirm either from the vendor, so treat them as folklore.
nOps splits the difference in a way that is easy to defend internally: visibility as a fee, and rate and commitment optimization billed as a percentage of realized savings, so no savings means no fee. Do the arithmetic before you sign, because a percentage cut of savings on a large enough bill quietly becomes larger than any flat SaaS fee, and how “savings” gets measured is exactly the sort of thing that belongs in the contract rather than the demo.
Picking, in two questions
Do you want to understand the bill or shrink it? Understanding is Infracost, Vantage, CloudZero, OpenCost, Kubecost. Shrinking without hiring someone is CAST AI, ScaleOps, or the nOps autopilot. Tools that promise both tend to be mediocre at one.
One cluster or a fleet? A single cluster with a budget-conscious team: OpenCost, and stop there until it hurts. A few clusters that need reconciled numbers and access control: the Kubecost free tier until retention or RBAC forces the conversation. Multi-cloud with Kubernetes as one slice: Vantage if you want prices you can read, CloudZero if finance is asking customer-level margin questions. Large cluster spend where the waste is real money and you are willing to let automation touch production: the autonomous tier.
Underneath all of it, none of these tools justify themselves until the platform fee disappears into the noise of the bill it is watching. Below roughly $50k a month, cloud-native tools plus Infracost will get you most of the way.
What none of them fix
Attribution of model API spend to an end user still comes apart everywhere. The tools can split by API key. Mapping key to user needs metadata you have to send, and almost nobody sends it cleanly.
Multi-cloud egress stays under-reported, because the underlying bills make it hard and the tools inherit that. Forecasting is a regression on last quarter wearing a nicer chart. And no tool fixes a tagging culture: they make one survivable, which is not the same thing.
One thing genuinely moving underneath: the FinOps Foundation’s FOCUS specification is on 1.4, and the shared-cost and split-allocation fields it added are aimed squarely at the allocation problem this whole category was invented for. Native billing exports are slowly getting better at the job. Not soon enough to change your decision this quarter, but if you are picking something for a five-year horizon, favor the ones that speak it.
Deploy OpenCost this week and look at where the money goes, remembering that the numbers are list price and therefore high. That one afternoon tells you which problem you have, and you cannot pick a tool before you know whether it is a visibility problem or a waste problem.