Cloud Egress Costs in India: AWS, Azure, Google Cloud and Indian Providers Compared

Cloud infrastructure pricing is usually discussed in this order:

Compute → Storage → Database → Support

Network transfer often appears later.

That can be expensive.

A workload may look economical when it is tested with a few users and become much more costly once it begins:

• Serving video
• Delivering customer downloads
• Returning AI-generated media
• Moving model outputs
• Restoring backups
• Synchronizing data between clouds
• Sending large API responses
• Replicating data across regions

For bandwidth-heavy workloads, egress can become one of the largest variable lines on the cloud bill.

The useful equation is:

This is why cloud providers should not be compared using VM price alone.

For a broader view of secondary cloud charges, see the Cloud Pricing Hidden Costs in India guide.

What Is Cloud Egress?

Cloud egress is data leaving a cloud service or cloud network.

The most familiar example is:

Cloud VM → Public Internet

But cloud networking can create several different transfer charges.

These are not interchangeable.

A provider may advertise free inbound traffic while charging outbound traffic.

A service may offer free same-region transfer but charge when traffic crosses Availability Zones.

A CDN may use a different rate card from compute instances.

A NAT gateway can add its own per-GB processing fee before the normal internet-egress charge is applied.

The first step is therefore to map where the bytes actually travel.

Cloud Egress Pricing in India at a Glance

The following table compares common public internet-transfer models.

These are not perfectly equivalent services, and the Indian-provider terms can be SKU-specific.

The biggest lesson is not that Indian providers are always cheaper.

It is that the billing model is different enough to change the architecture economics.

AWS Egress: Familiar, but Region Pricing Matters

AWS gives customers 100 GB per month of internet data transfer out at no charge, aggregated across eligible AWS services and regions except China and GovCloud.

That allowance includes services such as Amazon EC2, Amazon S3, Amazon RDS, Elastic Load Balancing, Amazon SageMaker and CloudWatch Logs.

After the free allowance, internet data transfer is charged using regional tiers.

For Mumbai, the first paid tier has historically been around the $0.109/GB range, with lower rates as monthly transfer increases.

Because AWS pricing can differ by source region and service, buyers should verify the current Mumbai SKU in the AWS Pricing Calculator before committing a production forecast.

What Does 10 TB of AWS Egress Mean?

At approximately $0.109 per GB after the first 100 GB, a 10 TB/month workload can create a network line item of roughly $1,080 before tax, depending on whether the workload is measured in GB or GiB and which AWS service/rate applies.

That is before other networking components such as:

• NAT Gateway
• Cross-AZ transfer
• Inter-region transfer
• Transit Gateway
• Global Accelerator
• CloudFront

The important question is not:

What is the EC2 hourly rate?

It is:

What does the complete request path cost after users start moving data?

AWS NAT Gateway Can Stack Another Per-GB Charge

AWS NAT Gateway has its own billing model.

AWS documents that a NAT gateway can generate:

• NAT Gateway hourly charges
• NAT Gateway data-processing charges
• Standard data-transfer charges

That means internet-bound traffic from a private subnet can incur both:

NAT processing + Internet data transfer

depending on the architecture.

AWS also warns that cross-Availability-Zone paths to a NAT gateway can add additional transfer charges.

AWS explicitly recommends endpoint-based architectures where appropriate as one way to reduce NAT Gateway cost.

Azure Internet Egress From India

Azure publishes internet egress rates by source continent.

For traffic from Asia, excluding China, routed over Microsoft’s Premium Global Network, the current public pricing is:

Azure defines 1 TB as 1,000 GB in this table.

So a workload sending 10 TB/month from an India Azure region over the Premium Global Network would have a simplified outbound calculation of:

9,900 GB × $0.12 = approximately $1,188/month

before tax and any additional networking services.

Azure Has a Lower-Cost Routing Preference

Azure also publishes a separate rate for internet traffic routed through transit ISP networks rather than its Premium Global Network.

For Asia:

At 10 TB/month, the simplified first-tier calculation becomes:

9,900 GB × $0.11 = approximately $1,089/month

That is about $99/month less than the Premium Global Network example at the same 10 TB volume.

The trade-off is routing behavior.

Lower transfer pricing is useful only if the network performance still meets the workload’s requirements.

Azure NAT Gateway Adds Another Layer

Azure NAT Gateway can also generate its own charges.

Microsoft’s pricing documentation states that NAT Gateway is billed hourly, data processed is billed per GB, and standard Azure bandwidth charges can still apply in addition to NAT processing.

That means the same general warning applies as on AWS:

Outbound bandwidth price is not always the complete networking price.

When the architecture uses private subnets, NAT, firewalls, load balancers or other data-processing services, calculate the full path.

Google Cloud Premium Tier

Google Cloud’s Premium Tier routes traffic using Google’s global network for a larger part of the path.

Internet data-transfer pricing depends on source region, destination geography, usage volume and product.

For traffic from India to many destinations in Asia, Google currently lists Premium Tier internet data transfer starting around:

• First applicable free usage
• $0.12/GiB for the initial paid tier
• $0.11/GiB at the next tier
• Lower rates at larger volumes

This is not directly comparable with Azure’s GB-based table because Google frequently prices network transfer in GiB.

Do not mix GB and GiB when modelling larger workloads.

Google Cloud Standard Tier Is the More Interesting Cost Option

Google Cloud Standard Tier uses a more public-internet-oriented routing model.

For both Mumbai (asia-south1) and Delhi (asia-south2), Google currently publishes:

At a full 10 TiB of monthly Standard Tier transfer:

• Total transfer: 10,240 GiB
• Free allowance: 200 GiB
• Billable: 10,040 GiB
• Rate: $0.085/GiB

Simplified network cost:

10,040 × $0.085 = approximately $853.40/month

That is materially lower than Azure’s Premium Global Network example at a broadly similar traffic scale.

But it is not a pure price comparison.

Standard Tier and Premium Tier are intentionally different network products.

Hyperscaler Egress Cost at Roughly 10 TB

The following examples are useful for understanding scale, but they are not perfectly apples-to-apples.

Azure uses decimal GB/TB in its published table. Google Standard Tier uses GiB/TiB. AWS is region-specific.

Do not use this table as a final provider ranking.

It excludes NAT, CDN, cross-zone traffic, inter-region traffic, taxes, private connectivity, negotiated rates, support, and differences in network behavior.

Its purpose is to show why egress deserves its own line in the cloud model.

AceCloud: Unmetered for Most Standard Workloads

AceCloud’s current public pricing FAQ says it offers unmetered bandwidth and does not charge separately for data transfers under normal usage.

Its more detailed FAQ adds an important qualification.

AceCloud states that bandwidth is unmetered for most workloads, while sustained CDN-scale traffic in the tens of terabytes per month from a single instance may be discussed during onboarding.

For India-first workloads, this model can materially change total cost compared with clouds that charge every outbound GB.

It still needs to be evaluated alongside compute price, port throughput, region, SLA, support, storage, GPU availability and contract terms.

Yotta Shakti: Transfer Terms Depend on the SKU

Yotta Shakti is a good example of why cloud egress comparisons need service-level detail.

Shakti’s public pricing pages currently show unlimited ingress and egress included with several AI Workspace, AI Lab, GPU and cluster configurations.

However, the current Shakti pricing page also lists generic networking add-ons including:

• Ingress: ₹5.99/GB
• Egress: ₹5.99/GB

This means it would be inaccurate to say:

Yotta Shakti has zero egress charges.

The safer statement is:

Several Shakti AI/GPU SKUs include unlimited data transfer, while transfer can be separately priced for other services or add-ons.

Buyers should therefore confirm the specific SKU in the commercial proposal.

E2E Networks: Object Storage Has a Clear External Transfer Rate

E2E’s current Object Storage documentation gives a more explicit service-specific example.

For E2E Object Storage (EOS):

Data transferred from E2E to external/public networks: ₹3/GB

E2E notes that this charge applies specifically to external networks.

The same documentation says traffic between E2E Compute Nodes and E2E Object Storage can use up to 40 Gbps, although actual performance depends on factors such as object size, compute resources and network path.

This is useful pricing clarity, but it should not be generalized to every E2E product.

The E2E Networks provider page can be used for the broader provider comparison.

Utho: Public Pricing Needs Clarification Before Procurement

Utho’s current website contains two different ways of describing bandwidth.

Its Simple Pricing page says customers will not see hidden bandwidth charges, bandwidth overage charges or egress fees.

However, Utho’s main compute pricing page lists 1,000 GB transfer with many cloud-instance plans.

Those two messages are not necessarily impossible to reconcile, but the public pages do not make the relationship completely clear.

For procurement, ask Utho to confirm in writing:

1. Whether outbound traffic is capped, throttled or merely measured after 1 TB.
2. Whether overage after the included transfer is charged.
3. Whether object storage follows the same network policy as compute.
4. Whether the rule varies by plan or region.
5. Whether sustained high-bandwidth traffic is subject to fair-use terms.

The signed commercial terms should take priority over a general marketing statement.

Why “Zero Egress” Is Not the Same as “Free Networking”

A provider can charge no per-GB egress fee and still have networking costs.

Possible charges include:

Likewise, an unmetered bandwidth plan may still have port-speed limits, fair-use terms, sustained-transfer policies or service-specific exclusions.

The better procurement question is:

What is the total monthly network cost at our expected throughput and traffic volume?

Five Egress Scenarios Worth Modelling

The same cloud can be cheap for one workload and expensive for another.

1. SaaS API

Assume moderate response payloads, around 2 TB/month internet egress and mostly Indian users.

Main concerns:

• Internet transfer
• NAT
• Load balancing
• API architecture

2. Video or Media Delivery

Assume 20 TB/month outbound, large files and repeated customer downloads.

Main concerns:

• Egress
• CDN
• Cache hit ratio
• Port throughput

A zero-egress or bundled-bandwidth model can become much more financially significant here.

3. AI Inference

Text-only inference may not create enormous egress.

Image, audio and video AI can.

Main concerns include output size, API response volume, model downloads, dataset movement and cross-cloud pipelines.

4. Backup and Disaster Recovery

Normal monthly egress may be low.

A real recovery can suddenly move tens of terabytes.

Main concerns:

• Retrieval
• Egress
• Restore throughput
• Cross-region recovery
• Recovery time

5. Multi-Cloud Architecture

A workload continuously moving data between providers can create egress on one or both sides of the architecture.

Main concerns:

• Cloud-to-cloud transfer
• Inter-region charges
• Private connectivity
• Database replication
• Observability traffic

Multi-cloud should therefore not be assumed to reduce cost.

Exit Egress Is Different From Everyday Application Egress

There is an important exception to normal hyperscaler egress pricing.

AWS and Azure both publish programs that can waive eligible internet data-transfer charges when customers are leaving the platform.

AWS

AWS says eligible customers can request free data transfer out when moving all data off AWS or all data off a particular AWS service, subject to its process and conditions.

The transfer has to be part of a genuine move rather than normal application traffic.

Azure

Azure also offers eligible customers free egress when leaving Azure for another cloud provider or an on-premises data centre.

Customers need to follow Microsoft’s support and credit process.

This distinction matters:

Normal customer traffic ≠ Exit migration traffic

A workload serving 10 TB/month to users can still pay standard egress even though a one-time provider-exit transfer may qualify for a waiver.

CDN Can Change the Egress Equation

Before moving workloads solely because origin egress looks expensive, model CDN architecture.

A CDN can:

• Cache frequently requested objects
• Reduce repeated origin traffic
• Move delivery closer to users
• Improve latency
• Use a different bandwidth rate card

But the CDN is not automatically cheaper.

Calculate:

Origin-to-CDN Transfer + CDN Requests + CDN Internet Delivery + Cache Misses

AWS, for example, states that data transfer from AWS services to Amazon CloudFront is free, while CloudFront has its own delivery pricing.

For media and content workloads, the more useful metric is often:

Cost per TB successfully delivered to users

rather than raw origin egress price.

Egress Can Also Affect Cloud Architecture

High outbound cost can influence technical decisions.

Teams may choose to:

• Cache more aggressively
• Compress API responses
• Keep processing close to stored data
• Avoid unnecessary cross-region copies
• Use private endpoints instead of NAT
• Place analytics where the dataset already lives
• Reduce verbose telemetry
• Batch transfers
• Use a CDN
• Avoid moving large AI datasets repeatedly

This is why egress is not only a finance problem.

It is an architecture variable.

How to Estimate Your Monthly Egress Bill

Start by mapping every traffic path.

Step 1: Estimate Internet Outbound

Record monthly GB/TB, destination geography, source region and expected growth.

Step 2: Add NAT Processing

For private workloads, determine whether traffic passes through a managed NAT service.

Step 3: Add Cross-Zone Traffic

Check whether load balancers, databases, Kubernetes nodes or NAT gateways create cross-zone transfer.

Step 4: Add Inter-Region Traffic

Include disaster recovery, replication, backups, multi-region databases and log aggregation.

Step 5: Add CDN

Estimate cache hit ratio, origin traffic, CDN requests and CDN outbound volume.

Step 6: Add Tax and Currency

For Indian buyers, compare the final invoice.

USD pricing can introduce:

• Exchange-rate movement
• Card/forex costs
• GST treatment
• Different billing entities

The INR vs USD Cloud Billing Guide covers those differences in more detail.

A Better Cloud-Egress Comparison Checklist

Before choosing a provider, ask:

1. How much internet egress is included each month?
2. What does the first paid tier cost?
3. Does pricing use GB or GiB?
4. Does the rate change by region or destination?
5. Is there a separate low-cost routing tier?
6. Is NAT charged per hour and per GB?
7. Is cross-AZ traffic charged?
8. Is same-region private transfer free?
9. What does inter-region replication cost?
10. Is origin-to-CDN transfer charged?
11. Does object storage have separate egress pricing?
12. Is bandwidth bundled with the instance?
13. Does “unmetered” have fair-use conditions?
14. Is the port speed sufficient at peak load?
15. What happens above the included transfer allowance?
16. Are there different rates for GPU and CPU services?
17. Can volume pricing be negotiated?
18. What taxes apply to the final invoice?
19. Is provider-exit migration eligible for an egress waiver?
20. What would a 10 TB, 50 TB and 100 TB month cost?

The Cloud Provider Selection Checklist for India can be used for the wider infrastructure evaluation.

When Does a Bundled or Unmetered Model Matter Most?

Bundled bandwidth becomes particularly valuable when outbound traffic is both large and predictable.

That does not mean a bandwidth-heavy workload should automatically leave a hyperscaler.

Hyperscalers may still justify higher network cost through global regions, managed databases, security tooling, edge infrastructure, enterprise support, managed AI services and existing engineering skills.

The correct metric is cost per successful workload, not the cheapest egress line in isolation.

FAQs

Final Takeaway

Do not compare cloud providers using VM price alone.

For bandwidth-heavy applications, compare:

Compute + Storage + Internet Egress + NAT + Cross-Zone Transfer + Inter-Region Transfer + CDN + Tax

AWS, Azure and Google Cloud provide deep ecosystems and mature global networking, but their standard internet-transfer models can become expensive at multi-terabyte scale.

Indian providers are introducing different economics.

AceCloud advertises unmetered bandwidth for most standard workloads. Selected Yotta Shakti AI/GPU services include unlimited transfer. E2E provides an explicit ₹3/GB external rate for object storage. Utho advertises no egress fees but should clarify how that promise relates to the transfer allowances shown on its compute plans.

No single model wins every workload.

For a small internal application, egress may barely matter.

For media, backups, AI data pipelines or SaaS products sending tens of terabytes per month, it can change the provider decision entirely.

Price the traffic pattern you expect in production, not the empty architecture diagram.

Use the getInfra.cloud provider comparison hub alongside your own bandwidth model before making the final infrastructure decision.

Sources and Pricing References

Provider rates, allowances, fair-use terms and transfer policies can change. Verify the current service-specific pricing and commercial terms before making a production purchasing decision.