When you open your AWS billing dashboard, find the line item labeled "Data Transfer Out." For most teams running active data workloads: AI pipelines, analytics exports, backup restore tests, that line is bigger than the storage line. Often much bigger. That cost has a name: egress fees.
This guide covers exactly what egress fees are, how AWS, Azure, and GCP calculate them in 2026, what they cost at three realistic workload scales, and what your options are for eliminating them.
What Are Egress Fees?
Egress fees are charges that cloud providers bill when data leaves their network, when you download, transfer, or serve data from cloud storage to the internet, to another region, or to another provider. Uploading data (ingress) is free on all major platforms. Downloading is not.
AWS S3 charges $0.09/GB for the first 10TB of internet egress each month. Azure Blob Storage charges $0.087/GB. Google Cloud Storage charges $0.12/GB for the first terabyte. These rates apply each time data crosses the billing boundary, every model training run, every analytics export, every backup restore test, every API response that returns stored data.
The defining characteristic of egress fees is the asymmetry: cloud providers charge 4–6× more to retrieve data than to store it. Azure charges $0.018/GB/month to store data and $0.087/GB to download it - a 4.8× multiplier. That gap is not incidental. It is the mechanism that makes migrating away from a cloud provider expensive, regardless of what any service agreement says about portability.
How Egress Fees Are Calculated?
Cloud egress fees follow a tiered, per-gigabyte model. The exact triggers vary by provider, but the core formula is consistent: volume of data transferred (GB) × applicable rate per GB.
What Triggers an Egress Charge
- Data downloaded from cloud storage to the internet (browsers, API clients, on-premises servers)
- Data transferred between cloud regions (us-east-1 → eu-west-1)
- Data transferred to a different cloud provider (AWS → GCP)
- CDN origin pulls in some configurations
What Typically Does Not Trigger an Egress Charge
- Data transferred between services within the same region and provider (EC2 → S3 in the same AWS region is usually free)
- Inbound data, uploads are almost universally free on all major providers
A Simple Calculation
A team running a 50TB analytics dataset on AWS S3, serving 20TB of data per month to ML pipelines and downstream users, pays:
- First 10TB (10,000GB) at $0.09/GB = $900
- Next 10TB (10,000GB) at $0.085/GB = $850
- Total monthly egress: $1,750
Storage for that same 50TB runs approximately $1,150/month. The egress line is 60% of the total bill. The "cheap cloud storage" costs nearly three times what the storage rate implies when active workloads are involved.
2026 Egress Fee Tables: AWS, Azure, and GCP
Prices verified against vendor pricing pages as of January–April 2026. Cloud providers update rates without notice, verify your configuration at each provider's live pricing calculator before budgeting.
AWS S3, Data Transfer Out to the Internet
Free tier: 100 GB/month aggregated across all AWS services.
Storage rate (Standard, us-east-1): ~$0.023/GB/month, verify at aws.amazon.com/s3/pricing before budgeting, as regional rates vary.
Source: aws.amazon.com/s3/pricing/
Azure Blob Storage, Bandwidth (Zone 1: North America, Europe, Asia Pacific)
Free tier: 100 GB/month.
Storage rate (Hot, LRS): $0.018/GB/month.
Source: azure.microsoft.com/en-us/pricing/details/bandwidth/
Google Cloud Storage, Internet Egress (Premium Tier, North America origin)
Additional retrieval fees apply on top of egress for Nearline ($0.01/GB), Coldline ($0.02/GB), and Archive ($0.05/GB) storage classes.
Storage rate (Standard): $0.020/GB/month.
Source: cloud.google.com/storage/pricing
Egress-to-Storage Multiplier: The Core Problem
Egress consistently costs more, often far more, than one full month of storage for the same data volume. For teams that access their data regularly, the storage rate is the least important number on the invoice.
3 Real-World Cost Scenarios
Scenario 1: Series A AI Startup, 20TB Stored, 15TB Monthly Egress
A team building an AI product: training data, model checkpoints, inference outputs. Models retrain weekly against stored datasets. Production API serves model outputs to end users. This is a standard AI engineering workload.
Egress math:
- AWS: 10,000GB × $0.09 = $900; 5,000GB × $0.085 = $425. Total: $1,325
- Azure: 15,000GB × $0.087 ≈ $1,305 (100GB free tier simplified out)
- GCP: 1,000GB × $0.12 + 9,000GB × $0.11 + 5,000GB × $0.08 = $120 + $990 + $400 = $1,510
Egress represents 74% of the AWS bill, 78% of the Azure bill, 79% of the GCP bill. Annual savings vs. Azure: ~$16,464. At Series A scale, that difference is runway, not just a line item on a cloud cost dashboard.
Scenario 2: Growth-Stage Analytics Platform, 75TB Stored, 40TB Monthly Egress
A data platform serving analytics to internal business users, data science teams, and external partners. Data is queried daily. Exports, scheduled reports, and partner data feeds drive egress volume. Egress scales with the product's success.
Egress math:
- AWS: 10,000GB × $0.09 + 30,000GB × $0.085 = $900 + $2,550 = $3,450
- Azure: 40,000GB × $0.087 = $3,480
- GCP: 1,000GB × $0.12 + 9,000GB × $0.11 + 30,000GB × $0.08 = $120 + $990 + $2,400 = $3,510
Annual savings vs. Azure: ~$44,784. At this scale, egress costs exceed what a full-time data engineer costs to hire. Teams optimizing "cloud spend" while paying $4,830/month for a 75TB dataset are optimizing the wrong number.
Scenario 3: Enterprise DR Testing, 100TB Stored, 10TB Monthly Restore Testing
An enterprise maintaining backup storage for disaster recovery and compliance. Ransomware preparedness frameworks (NIST CSF, most cyber insurance requirements) mandate documented restore testing. Every restore test downloads data from backup storage, and every download triggers an egress charge.
Honest caveat: For this scenario, Azure Cool and GCP Nearline have lower headline storage rates ($0.01/GB) than Akave's flat rate. On a pure storage-only comparison, data you almost never retrieve, hyperscaler archive tiers can be cheaper. Akave's advantage is the egress line: the moment restore testing volume increases, or quarterly full-restore validation is required by compliance frameworks, the math shifts decisively. A full 100TB restore on Azure Cool costs $8,700 in egress alone. On Akave: $0.
How to Eliminate Egress Fees from Storage Cost?
There are three substantive approaches. Each has a different ceiling.
Approach 1: Optimize Within Your Current Provider
Reduce data crossing billing boundaries through caching, CDN for static assets, co-locating compute and storage in the same region, and batching API calls.
The ceiling: Effective for web assets and read-heavy static content. Has near-zero impact on AI training pipelines, analytics exports, or backup restore operations, these workloads need to move data. Reducing egress on an AI training pipeline means training less frequently or on smaller datasets. Neither is a viable outcome for a data engineering team under pressure to ship.
Approach 2: Negotiate Enterprise Discounts or Private Interconnects
AWS Direct Connect, Azure ExpressRoute, and Google Cloud Interconnect can reduce per-GB rates for specific traffic patterns. Enterprise agreements unlock committed-use discounts at sufficient volume.
The ceiling: Private interconnects add infrastructure costs, provisioning, maintenance, colocation. Committed-use discounts reduce flexibility. You are still paying for egress at a lower rate. The underlying model, charging you to access your own data, doesn't change.
Approach 3: Move Egress-Heavy Workloads to a Zero-Egress Provider
For workloads where egress represents 50%+ of total cloud spend, switching to a provider with a zero-egress model is the only approach that eliminates the cost driver rather than managing it.
Akave Cloud charges $14.99/TiB/month for object storage with $0 egress fees within transparent fair-use thresholds designed for production workloads, AI training, analytics pipelines, backup restore testing, external data serving. The S3-compatible API means standard object storage workflows (PUT, GET, DELETE, LIST, multipart uploads) work without application-layer changes for most teams. Workloads using lifecycle policies, object versioning, Lambda-equivalent triggers, or KMS-managed key patterns should test their specific configuration before migrating.
When Akave fits:
- Egress represents 50%+ of your current storage bill
- You serve data externally to users, partners, or production APIs
- You train AI/ML models against stored datasets on a regular schedule
- You run mandatory backup restore testing or quarterly DR validation
- Budget predictability matters to a finance team building annual cloud cost forecasts
When it doesn't fit:
- Pure archival with near-zero retrieval (hyperscaler archive tiers are cheaper for this specific pattern)
- Workloads deeply integrated with hyperscaler-native services (Lambda triggers, KMS-managed keys, proprietary event notification systems)
- Compliance frameworks that mandate a specific hyperscaler certification or dedicated tenancy model
FAQ
What are egress fees?
Egress fees are per-gigabyte charges billed by cloud providers when data leaves their network. This includes downloading data to the internet, transferring between regions, or moving data to another provider. AWS S3 charges $0.09/GB for the first 10TB of monthly internet egress. Azure charges $0.087/GB. Google Cloud charges $0.12/GB for the first 1TB. Uploading data (ingress) is free on all major platforms.
How are cloud egress fees calculated?
Egress fees are calculated as data volume (in GB) multiplied by the provider's per-GB rate for the applicable traffic type and volume tier. All three major providers use tiered pricing, the per-GB rate decreases as monthly volume increases. A team transferring 15TB/month from AWS S3 pays $0.09/GB for the first 10TB and $0.085/GB for the next 5TB, totaling $1,325. This is typically more than the monthly storage cost for the same data.
Why are egress fees 4–6× the cost of storage?
Cloud providers charge 4–6× more for egress than storage because data gravity creates switching costs. Once 100TB lives on AWS, downloading it to move elsewhere costs $8,700 in egress fees before you pay your new provider a dollar. The pricing multiplier reflects market position: storage attracts customers, and egress pricing retains them. The 4.8× Azure multiplier and 6× GCP multiplier exceed actual bandwidth economics by a significant margin.
How does Akave compare to AWS and Azure for total storage costs?
For workloads with active egress, Akave Cloud consistently produces lower total cost of ownership. A Series A startup with 20TB stored and 15TB monthly egress pays $1,665/month on Azure and approximately $293/month on Akave, a difference of $16,464/year. A growth-stage analytics platform with 75TB stored and 40TB monthly egress pays $4,830/month on Azure and approximately $1,098/month on Akave, a difference of $44,784/year. Akave charges $14.99/TiB/month for storage with $0 egress fees under production-grade fair-use thresholds.
Are there workloads where AWS or Azure is still the right choice?
Yes. For pure archival storage with near-zero retrieval, hyperscaler archive tiers (S3 Glacier, Azure Archive, GCP Archive) have lower storage rates than Akave's flat rate, and the egress savings don't materialize if data is rarely accessed. Akave's advantage scales directly with egress volume. If egress is a small fraction of your bill and your workload is mostly cold storage, the flat-rate model may not deliver meaningful savings. The breakeven point varies by workload, run the numbers against your specific egress patterns.
How long does migration from S3 to Akave take?
For standard S3 object-storage patterns, PUT, GET, DELETE, LIST, multipart uploads, the primary change is updating the endpoint URL and rotating credentials. Akave's S3-compatible API means application-layer code typically requires no changes for these patterns. Teams running straightforward object storage workloads have completed endpoint switches in under a day. Workloads using S3-specific features such as Lambda triggers, KMS-managed encryption keys, lifecycle policies, object versioning, or event notifications require additional planning before migrating. Test your specific configuration before committing to a full migration.
If my cloud provider offers enterprise egress discounts, why switch?
Enterprise discounts reduce the per-GB rate but don't change the model, you're still paying to access your own data. A 20% enterprise discount on Azure egress saves $261/month on a $1,305/month egress bill for a 15TB/month workload. That discounted Azure bill still costs more than Akave's full monthly invoice for the same workload. Negotiated discounts also require volume commitments that reduce flexibility. For egress-heavy workloads, eliminating the cost driver produces larger savings than optimizing the rate.
Does Akave's zero-egress model have usage limits?
Akave's $0 egress applies within fair-use thresholds designed for production workloads, AI training, analytics pipelines, backup restore testing, and external data serving. Current thresholds are published at akave.com/pricing. For workloads that exceed published thresholds, Akave offers enterprise terms. The practical point: Akave's thresholds are calibrated for normal production access patterns, not consumer-grade use. You won't hit them running standard data engineering or AI workloads, but verify your specific pattern against the published limits before migrating.
Start Eliminating Egress Costs
Substituting your own storage and egress numbers into the scenarios above takes two minutes. If egress is more than half your storage bill, the math will make the next step obvious.
- Run side-by-side cost comparisons at akave.com/pricing
- Start a free trial at akave.com/free-trial
- See how Akave integrates with Snowflake external volumes at akave.com/solutions/snowflake
Further Reading
- Cloud Egress Fees Explained: Azure vs Google Cloud vs Akave, Deep-dive comparison with workload scenarios at 50TB, 1PB, and 10PB scale
- Cloudflare R2 vs AWS S3 vs Akave: Best Zero-Egress Storage for AI, How Akave compares to zero-egress alternatives including Cloudflare R2
- AI/ML Data Lake solutions, Storage architecture for ML teams with regular training cycles and model serving