CAST AI Pricing: What CAST AI Costs, What Is Free, and How It Compares to Kubecost

CAST AI does not publish a price. As of 9 September 2026 the pricing page at cast.ai/pricing is a quote request form headed "Get a custom quote", and it says plainly that "Cast AI's pricing model depends on a few factors that are specific to your environment, so we'll need to get in touch to give accurate pricing information." There is no tier table, no per-CPU rate card and no stated trial length. Every specific CAST AI dollar figure ranking in search results today comes from a competitor blog, a reseller marketplace or a software-review aggregator, not from CAST AI.

That matters more than it sounds, because the figures circulating are precise enough to look authoritative. You will see a Growth plan quoted at $1,000 a month plus $5 per CPU per month, and you will see the model described as a percentage of realized savings plus a per-CPU fee. Neither appears on CAST AI's own site. Budgeting against them, or worse, walking into a negotiation anchored on them, is how teams end up surprised by the actual quote.

Everything below was read off vendor pages on 9 September 2026, or is a dated public announcement. Where a vendor publishes nothing, this page says so rather than repeating a number from someone selling against them.

How much does CAST AI cost?

Nobody outside CAST AI and its customers knows, and the vendor has made that a deliberate choice. The pricing page collects your details and routes you to sales, with the explicit reasoning that the model depends on factors specific to your environment. In practice, for a Kubernetes automation platform, that means your quote will be driven by cluster size, the number of clusters, which capabilities you turn on, and your contract length. What you can plan around is the shape of the deal rather than its size.

The important consequence for your budget request is that the cost is not fixed against your headcount or your seat count. It scales with your infrastructure. That is the opposite of how most engineering tools are priced, and it is the single thing to model before you commit, because a platform fee that grows with your cluster can eat a meaningful share of the savings it delivers once the easy waste is gone.

Does CAST AI offer a free tier or a free trial?

There is a free entry point, and its boundaries are not published. The product pages carry "Start free trial" and "Start free" calls to action in several places, so you can begin without a contract. What the site does not state anywhere is the trial length, the cluster or vCPU ceiling on the free tier, or which capabilities are gated behind a paid plan. If free monitoring is what you want, get those three limits in writing before you connect a production cluster, because the difference between free read-only visibility and a trial of full automation is enormous from a change-management perspective.

Why do the CAST AI prices you find online disagree with each other?

Because none of the sources publishing them can see the rate card. Three kinds of site dominate these search results and each has a structural reason to be wrong. Reseller and marketplace sites publish medians of contracts they brokered, which reflect a self-selected slice of deals and go stale quickly. Competitor blogs publish figures chosen to make a comparison land a particular way, and they have no obligation to update them. Review aggregators repeat whatever a vendor's marketing page said at some point, often years ago. The tell is disagreement: when four sites quote four different structures for the same product, none of them is reading a rate card.

This is the same pattern that plays out across this category. It shows up in Kubecost pricing, where a widely repeated per-node figure appears nowhere on the vendor's page, and in Flexera pricing, where third-party medians and percentage-of-spend claims conflict with each other and with the vendor's silence. Treat any unattributed number in this category as a rumor until a vendor page or a signed quote confirms it.

How does CAST AI pricing compare to Kubecost, PerfectScale and StormForge?

The useful comparison is not the price, since most of the category refuses to publish one. It is what each vendor is willing to tell you before you talk to sales, because that predicts how the negotiation will go. Read off each vendor's own site on 9 September 2026 and 4 September 2026:

PerfectScale is the only automated vendor in this group publishing a usable free entry with a stated ceiling, at 200 vCPU per month across unlimited clusters. Kubecost's Foundations tier is genuinely free with a published limit of 250 cores and fifteen days of metric retention, though note that kubecost.com/pricing now redirects to apptio.com/pricing following IBM's acquisition in September 2024, and no paid Kubecost tier publishes a price either. If your evaluation criterion is "let me see the numbers before I book a call", that is a two-vendor shortlist, and the rest of the category is a series of sales conversations.

What are you actually paying for with CAST AI?

This decides whether the quote is reasonable, so it is worth being precise about scope. CAST AI is an automation platform rather than a reporting tool. Its published capabilities include a cluster autoscaler that provisions instances and scales on real-time demand, container live migration that moves running workloads between nodes, bin packing to consolidate workloads onto fewer nodes, a rebalancer that optimizes cluster state on a schedule or on demand, spot instance automation covering the interruption lifecycle, commitments utilization across commitment types, pod mutations, memory event handling that provisions resources when pods run out of memory, and advanced pod autoscaling covering both vertical and horizontal scaling.

That list is the answer to whether the price is worth it. Every one of those capabilities requires write access to your cluster and, in most deployments, to your cloud account. You are not buying a dashboard. You are buying an agent that will change what runs where, at three in the morning, without asking. Some teams want exactly that and get real savings from it. Others discover in a security review that the write access is the blocker, and that the reporting half of the value was all they needed. Establish which team you are before the quote, because the answer changes which vendors belong on your list at all. We cover the full field in our comparison of Kubernetes rightsizing tools, and the automation-first end of it specifically in ScaleOps alternatives.

Is a percentage of savings pricing model good for the buyer?

It sounds aligned and often is not, which is why you should establish the model before the number. The appeal is obvious: you pay only when the vendor delivers, so the downside looks capped. Three things complicate that in practice. First, the baseline is negotiable and it decides everything, so ask in writing how savings are measured, against what starting point, and for how long a resource counts as saved. Second, savings plateau. The first quarter of a rightsizing program removes obvious waste; year three removes very little, while the fee keeps being calculated. Third, if the fee is partly per-CPU, your bill grows as you scale even when the incremental savings do not.

The cleanest comparison is against a flat subscription. A flat fee is easy to budget, easy to defend at renewal and does not penalize growth. A savings-share fee lowers your first-year risk and raises your long-run cost if the tool works well. Neither is wrong, but they suit different situations, and you cannot compare a quote from one against a quote from the other without modeling three years. Do that model before you pick, not after.

What should I ask CAST AI before signing?

Six questions, in the order that matters. First, what is the billing metric, in writing: per vCPU, percentage of savings, flat platform fee, or a combination. Second, if any part is savings-based, how is the baseline established and how long does a saved resource keep counting. Third, what are the exact limits of the free tier and the trial, in vCPU or cores and in days. Fourth, what write permissions does the agent require in the cluster and in the cloud account, and can it run in a recommendation-only mode. Fifth, what is the term and the uplift cap at renewal. Sixth, what happens commercially if we shrink the cluster, since a metric tied to infrastructure should fall as well as rise.

The fourth question is the one that most often changes the outcome. If your security review will not approve cluster write access this quarter, an automation platform is the wrong purchase regardless of price, and you need visibility first. That sequencing question is the same one that governs how you provision and roll out the workloads in the first place, which is why teams that have their deployment and server provisioning under proper control usually find the cost conversation much simpler: they already know what is running and why.

The short version

CAST AI publishes no price, offers a free entry point with unpublished limits, and quotes against your environment. The specific figures you will find online are third-party and disagree with each other. Your leverage comes from scoping the capability list to what you will actually turn on, pinning the billing metric in writing, and modeling three years rather than one if any part of the fee is savings-based. If you need cluster automation, price it against StormForge, ScaleOps and Zesty on the same metric. If what you actually need is to see and allocate Kubernetes and cloud spend before anyone touches the cluster, that is a different and much faster purchase: connecting your accounts read-only gives you allocation, showback, anomaly detection and forecasting in a day, with no agent and no write access to approve.