21 Sep 2026
An AWS pricing calculator will tell you what your compute and storage cost. It will not tell you what your data transfer, your idle capacity, or your currency exposure cost, and for most companies running steady, predictable workloads, that is where the real bill hides.
An AWS pricing calculator will tell you what your compute and storage cost. It will not tell you what your data transfer, your idle capacity, or your currency exposure cost, and for most companies running steady, predictable workloads, that is where the real bill hides. A fair comparison between AWS and a private cloud has to include all of it, not just the line items AWS puts on the front page.
Compute and storage are the visible part of an AWS bill. Data movement is the part that surprises people.
As of 2026, AWS charges $0.09 per GB for the first 10 TB of internet egress each month, after a 100 GB free allowance, dropping to $0.085/GB for the next 40 TB, $0.07/GB for the next 100 TB, and $0.05/GB above 150 TB. That is before three charges most teams do not budget for separately: cross-Availability Zone traffic at $0.01/GB in each direction, NAT Gateway processing at $0.045/GB on top of standard egress, and a per-hour charge on every public IPv4 address, whether it is doing anything or sitting idle.
None of these show up prominently on the AWS pricing page for EC2 or S3. They show up on the bill. For a workload with a distributed, multi-AZ architecture, data transfer alone commonly runs 10 to 30 percent of total AWS spend, and in microservices-heavy setups it can climb higher still.
Flexera's 2026 State of the Cloud Report, based on a global survey of more than 750 cloud decision-makers, found that an estimated 29 percent of public cloud infrastructure spend goes to waste, the first increase in five years after a steady decline. Seventeen percent of organisations exceeded their public cloud budget in the past year, and 76 percent of large enterprises now spend more than five million dollars a month on public cloud alone.
That 29 percent is not a rounding error. On a company spending, say, ten lakh rupees a month on AWS, close to three lakh of that is going toward idle or overprovisioned capacity nobody is actively using. A private cloud, priced on committed capacity rather than metered consumption, removes most of the mechanism that produces that number in the first place: there is no idle-resource waste to measure when you are paying for a fixed footprint you actually sized.
Two documented cases are worth knowing, because they are not marketing claims, they are numbers companies published themselves.
Dropbox moved the majority of its infrastructure from public cloud to its own leased data centre facilities starting in 2016. By its own S-1 filing, the company saved nearly 75 million dollars over the following two years, and its gross margin rose from 33 percent to 67 percent over the same period, a change it attributed directly to the infrastructure move alongside revenue growth.
More recently, 37signals, the company behind Basecamp and HEY, documented its exit from AWS and Google Cloud in detail. In 2022 it was spending 3.2 million dollars a year across both clouds. It invested roughly 700,000 dollars in its own server hardware, and by 2024 had cut its cloud bill to 1.3 million dollars a year, saving nearly 2 million dollars in that year alone, with a public target of 7 to 10 million dollars in savings over five years as it finishes migrating off AWS S3 as well.
A widely cited 2021 analysis by the venture firm Andreessen Horowitz, “The Cost of Cloud, a Trillion Dollar Paradox,” argued that repatriation typically cuts infrastructure cost by roughly one-third to one-half once a company's workload has matured past its early, unpredictable growth phase. That report drew genuine pushback on how it translated cost savings into stock market impact, and that critique is fair. What was not seriously disputed, including by critics of the report, was the underlying cost data itself.
The pattern across all three cases is the same: none of these were early-stage startups guessing at demand. All three had reached a point where their workload was large enough, and steady enough, to know in advance roughly what capacity they needed.
Two costs are specific to running on AWS from India, and neither appears on any pricing page.
Two costs are specific to running on AWS from India, and neither appears on any pricing page.
Currency exposure. AWS bills in US dollars. A SaaS or enterprise business earning revenue in rupees but paying its infrastructure bill in dollars carries a currency risk that has nothing to do with engineering, and it moves independently of anything the company does well or badly.
Compliance retrofitting. SEBI's Framework for Adoption of Cloud Services by Regulated Entities requires data residency, customer-held encryption keys, and SOC-monitored deployments for regulated financial entities. Meeting these requirements on infrastructure not designed for them from the outset is a retrofit cost that a workload built on Indian-jurisdiction infrastructure from day one simply does not carry.
The economics change in three specific ways.
Egress stops being metered. A private cloud contract that includes bandwidth as part of a fixed monthly commitment removes the per-GB anxiety entirely. There is no calculator to run before shipping a large export to a customer.
Committed capacity is priced predictably. ZeaCloud's own commercial model, for example, prices short-term or no-commitment capacity at roughly an 18 percent premium over a committed rate, with that premium reaching zero at a 36-month commitment. The buyer decides, upfront, exactly what trade-off they are making between flexibility and cost, rather than discovering it after the fact on a bill.
The bill is in rupees. For an Indian company, this alone removes the currency variable from infrastructure planning.
None of this means every workload belongs on private infrastructure. AWS is genuinely the right choice for a company still finding product-market fit, where demand is unpredictable and the cost of guessing wrong on capacity is higher than the cost of paying a premium for elasticity. The venture analysis cited above makes the same point: cloud is close to unbeatable for a startup's first few years. The economics shift specifically once growth has become steady enough to plan around, which is usually well after that early phase.
Is AWS always more expensive than private cloud?
Not always. For unpredictable, early-stage, or highly bursty workloads, AWS's elasticity is worth its price premium. The cost gap opens up specifically for steady, predictable workloads where a company is effectively paying elasticity pricing for capacity it does not need to flex.
How much of an AWS bill is typically data transfer?
For distributed, multi-AZ architectures, data transfer commonly runs 10 to 30 percent of total AWS spend, and can run higher in microservices-heavy environments once cross-AZ traffic and NAT Gateway processing are included.
What is cloud repatriation?
It is the practice of moving workloads from public cloud back to owned or leased private infrastructure once a company's usage pattern has matured. Documented cases, including Dropbox and 37signals, show savings in the range of one-third to one-half of prior cloud spend.
Does moving off AWS mean giving up cloud entirely?
No. Most companies that move keep some workloads on public cloud, particularly for genuinely bursty or experimental workloads, while shifting steady-state, predictable capacity to private infrastructure priced on commitment rather than metering.
What is the biggest hidden AWS cost most companies miss?
Data transfer and idle capacity. Egress, cross-AZ traffic, and NAT Gateway fees rarely appear in initial cost estimates, and Flexera's 2026 research puts average wasted cloud spend at 29 percent industry-wide.
Disclaimer
ZeaCloud Services Private Limited operates ZeaStack, a managed sovereign private cloud platform priced on committed capacity from Tier III data centres in India. This article is part of ZeaCloud's ongoing coverage of cloud economics and FinOps.