On this page · 8 sections
Summary. Amazon EKS and Google GKE have converged on the same numbers: $0.10 per cluster per hour for a supported Kubernetes version and $0.60 once that version passes into extended support, a 6x step. On EKS the extended-support upgrade policy is set to EXTENDED by default for all new and existing clusters, and once a cluster has entered extended support you cannot switch it off. As of 16 August 2026 every EKS cluster still on Kubernetes 1.33 or older is already billing at the higher rate, and 1.34 has until 2 December 2026. Separately, AWS Lambda blocks function updates for nodejs20.x and python3.10 on 3 March 2027. The India premium is real but uneven: compute in Mumbai runs 5.21% above US East, S3 Standard 8.70% above, and data transfer out from 21.4% above at the first 10 TB to 60% above past 150 TB a month. Those five facts shape a cloud application budget far more than the framework choice does.
The cluster fee nobody notices until it doubles
Kubernetes upstream supports an active patch release series for roughly fourteen months: twelve months of standard patching, then a two-month maintenance mode in which release managers cut patches only for vulnerabilities with an assigned CVE ID and for dependency and base-image issues. Ordinary bug fixes stop. "Supported" in months thirteen and fourteen is a much thinner promise than in months one to twelve, and a lot of upgrade planning is built on not noticing that.
The managed services extend that runway and charge for it.
| Managed Kubernetes cost line | Amazon EKS | Google GKE |
|---|---|---|
| Cluster management, supported version | $0.10 per cluster per hour | $0.10 per cluster per hour, all modes and topologies |
| Extended support surcharge | $0.50 per cluster per hour | $0.50 per cluster per hour |
| Total on extended support | $0.60 per cluster per hour | $0.60 per cluster per hour |
| Extended-support runway | 12 months beyond standard support | Up to 24 months on the Extended release channel |
| Free allowance | None on the control plane | $74.40 in monthly credits per billing account, zonal and Autopilot clusters only |
| Default behaviour | Upgrade policy is EXTENDED by default on new and existing clusters | Extended channel is opt-in; no charge during standard support |
Read the default column twice. AWS documents it plainly: extended support is enabled by default, and for all new and existing clusters the upgrade policy is set to EXTENDED unless specified otherwise. Billing starts at the beginning of the day the version reaches end of standard support, in UTC.
The part that catches teams is the exit. AWS states that you cannot disable extended support once your cluster has entered it; you can only disable it while the cluster is still on standard support. Disabling it means the cluster is automatically upgraded to the next Kubernetes version when standard support ends. So the choice is a 6x fee or a forced upgrade on AWS's schedule, and it has to be made before the deadline, not after.
Google's free tier has a similar edge. The $74.40 monthly credit is described as equivalent to one free Autopilot or zonal Standard cluster, and it applies only to zonal and Autopilot clusters. A regional Standard cluster, which is what most teams run for high availability, pays from the first hour.
The EKS calendar, as it stands today
| Kubernetes version | End of standard support on EKS | End of extended support |
|---|---|---|
| 1.36 | 2 August 2027 | 2 August 2028 |
| 1.35 | 27 March 2027 | 27 March 2028 |
| 1.34 | 2 December 2026 | 2 December 2027 |
| 1.33 | 29 July 2026, already passed | 29 July 2027 |
| 1.32 | 23 March 2026, already passed | 23 March 2027 |
A single cluster left on 1.33 costs an extra $0.50 an hour, which is $360 a month, or $4,380 a year. That is trivial for one cluster and material across a fleet of twenty environments where the non-production clusters are the ones nobody upgrades. The upstream picture is tighter still: Kubernetes 1.34 enters upstream maintenance mode on 27 August 2026 and reaches end of life on 27 October 2026, so patching narrows well before the EKS billing change on 2 December.
The engineering judgement here is unglamorous. Cluster version drift is a finance problem before it is a security problem, and it is cheapest to fix on a schedule you set.
Serverless has its own wall
Teams that avoided Kubernetes by going serverless face a different date. AWS has bulk-aligned a large cohort of Lambda runtimes onto the same two deadlines, extending them beyond the usual 30 and 60 days after deprecation in response to customer feedback.
| Runtime | Deprecation | Block function create | Block function update |
|---|---|---|---|
| nodejs20.x | 30 April 2026, already deprecated | 1 February 2027 | 3 March 2027 |
| python3.10 | 31 October 2026 | 1 February 2027 | 3 March 2027 |
| python3.9 | in the same cohort | 1 February 2027 | 3 March 2027 |
| ruby3.2 | in the same cohort | 1 February 2027 | 3 March 2027 |
| provided.al2 | in the same cohort | 1 February 2027 | 3 March 2027 |
Amazon Linux 2 reached end of life on 30 June 2026, which is what pulled the python3.10, python3.11 and provided.al2 runtimes into this group. Blocked updates do not stop a function running. They stop you deploying a fix to it, which is a worse position to be in during an incident than an outage would be. The upgrade targets have their own clocks: nodejs22.x deprecates on 30 April 2027 and python3.11 on 30 June 2027, so a migration that lands on those buys about a year, while nodejs24.x and python3.12 buy considerably more.
What the India region actually costs
Both the AWS EC2 on-demand and S3 pricing pages render as JavaScript shells; the S3 page ships literal unsubstituted template tokens rather than numbers. The figures below come from the first-party pricing data those pages load at render time, and from Microsoft's own Azure Retail Prices API, because the Azure Linux VM pricing page is a shell too.
| Cost line | US East (N. Virginia) | Asia Pacific (Mumbai) | India premium |
|---|---|---|---|
| m6i.large on-demand, Linux | $0.0960 per hour | $0.1010 per hour | 5.21% |
| m7i.xlarge on-demand, Linux | $0.2016 per hour | $0.2121 per hour | 5.21% |
| t3.medium on-demand, Linux | $0.0416 per hour | $0.0448 per hour | 7.69% |
| S3 Standard, first 50 TB per month | $0.023 per GB | $0.025 per GB | 8.70% |
| Data transfer out, first 10 TB per month | $0.09 per GB | $0.1093 per GB | 21.4% |
| Data transfer out, above 150 TB per month | $0.05 per GB | $0.08 per GB | 60% |
Three observations that matter when you are choosing a region rather than complaining about one.
The compute premium is a flat multiplier, not a per-SKU markup. Every m-family size in Mumbai sits at exactly 5.208% above N. Virginia. Azure lands in the same place: a Standard_D2s_v5 Linux VM is $0.096 an hour in East US and $0.101 in Central India, also 5.21%, and within a tenth of a cent of AWS m6i.large in Mumbai for the same 2 vCPU and 8 GiB. Two vendors, one number. Region choice is not where you win on compute.
Egress is where the India penalty concentrates, and it widens with scale. At 21.4% for the first 10 TB it is an annoyance. At 60% above 150 TB a month it is an architectural constraint, and it is the reason a consumer-facing Indian product should be pushing traffic through a CDN and an origin shield rather than serving from the region directly. AWS gives 100 GB of free data transfer out per month, aggregated globally across all services and regions rather than per region.
The EKS control-plane fee, by contrast, is region-flat: $0.10 standard and $0.50 extended in Mumbai exactly as in N. Virginia. Data residency under the Digital Personal Data Protection Act, 2023 is therefore usually a cheaper decision than teams expect on compute and control plane, and an expensive one on egress.
How we scope a cloud build or re-platform
- Inventory the clocks first. Cluster versions against the EKS or GKE calendar, Lambda runtimes against the block-update dates, base images against their OS end-of-life. This produces a dated list, and the dated list is the roadmap.
- Separate the forced work from the wanted work. An upgrade you must do by 2 December 2026 is not a modernisation programme, and merging the two is how modernisation programmes slip.
- Model egress before choosing a region. On a consumer product the transfer-out line frequently exceeds the compute line, and it is the number with the widest India-versus-US gap.
- Set the cluster count deliberately. Extended-support surcharges are charged per cluster per hour, so a fleet of small per-team clusters carries a fixed cost that a smaller number of multi-tenant clusters with namespace isolation does not.
- Instrument cost per environment from the start. Non-production is where version drift and idle spend live, and neither is visible in an aggregate bill.
Our engagement model reflects that. A fixed-scope assessment produces the dated deprecation inventory, an egress-aware cost model for at least two region and delivery options, and a migration sequence that separates deadline-driven work from discretionary modernisation. The build or re-platform then runs on milestones tied to those dates, with a retained senior team afterwards, because the deprecation calendar does not stop when the project does.
eCorpIT builds and re-platforms cloud applications from Gurugram as an AWS, Microsoft and Google technology partner, with CMMI Level 5 assessed delivery processes. For related reading, our custom software development company guide covers scoping and pricing bespoke builds, DevOps services company covers the delivery pipeline and platform work, and SaaS development company covers multi-tenancy and billing for products you intend to sell.
FAQ
How eCorpIT can help
eCorpIT is a Gurugram-based engineering organisation, founded in 2021, that builds cloud-native applications and re-platforms existing ones on AWS, Azure and Google Cloud. We start with a dated deprecation inventory and an egress-aware cost model, so the migration sequence is driven by real deadlines and real unit costs rather than by a reference architecture. We design systems aligned with Digital Personal Data Protection Act, 2023 requirements and work as senior-led, multi-disciplinary teams under CMMI Level 5 assessed processes. Talk to our team if your cluster or runtime calendar needs sorting before it starts charging you.
References
Last updated: 16 August 2026.
