io.net charges zero egress fees for the first 1TB per month, then $0.05/GB for additional data transfer. This is 40-60% cheaper than AWS ($0.08-$0.12/GB), Azure ($0.087-$0.12/GB), and GCP ($0.08-$0.23/GB), which charge egress fees from the first gigabyte. For AI workloads that frequently download models, datasets, and training results, io.net's free 1TB allowance eliminates surprise bills. A typical ML team downloading 500GB/month pays $0 on io.net versus $40-60 on AWS, while teams moving 5TB/month save $250-550 monthly on data transfer costs alone.
Understanding Cloud Egress Fees
Egress (also called "data transfer out" or "bandwidth") refers to data leaving a cloud provider's network to the internet or another region.
Common Egress Scenarios in AI/ML:
- Downloading trained model weights (1-50GB per model)
- Exporting training results and logs (10-100GB per experiment)
- Transferring datasets for local analysis (100GB-10TB)
- Serving inference API responses to end users (varies by traffic)
- Syncing data to other cloud services or on-premise storage
Why Egress Fees Exist:
Traditional cloud providers charge egress fees to:
1. Recover network infrastructure costs
2. Incentivize keeping data within their ecosystem (lock-in)
3. Generate additional revenue (egress is 20-30% profit margin)
io.net's decentralized model has lower network overhead, passing savings to users.
Complete Egress Pricing Comparison
Real costs across major cloud GPU providers:
| Provider | Free Tier | First TB | Next 9TB | Next 40TB | 50TB+ |
|---|---|---|---|---|---|
| io.net | 1TB/month free | $0.00/GB | $0.05/GB | $0.05/GB | $0.04/GB (contact sales) |
| AWS | 100GB/month | $0.08/GB | $0.085/GB | $0.07/GB | $0.05/GB |
| Azure | None | $0.087/GB | $0.083/GB | $0.07/GB | $0.05/GB |
| Google Cloud | None | $0.12/GB | $0.11/GB | $0.08/GB | $0.08/GB |
| CoreWeave | 500GB/month | $0.05/GB | $0.05/GB | $0.05/GB | $0.04/GB |
| Lambda Labs | Unlimited* | $0.00/GB | $0.00/GB | $0.00/GB | $0.00/GB |
*Lambda Labs' unlimited egress applies only to their limited GPU inventory, which is frequently sold out.
Monthly Cost Examples:
Small AI Startup (500GB egress/month):
- io.net: $0 (within 1TB free tier)
- AWS: $32 (500GB × $0.08/GB after 100GB free)
- Savings: $32/month (100%)
Mid-Size ML Team (5TB egress/month):
- io.net: $200 ($0 for first 1TB + 4TB × $0.05/GB)
- AWS: $410 ($0.08/GB for first 10TB tier)
- Savings: $210/month (51%)
Enterprise AI Lab (50TB egress/month):
- io.net: $1,960 ($0 for first 1TB + 49TB × $0.04/GB custom rate)
- AWS: $3,400 (blended rate across tiers)
- Savings: $1,440/month (42%)
Hidden Egress Costs in AI Workflows
Data transfer fees accumulate from activities you might not expect:
1. Model Training Iterations:
Training a large language model involves multiple download/export cycles:
- Download pre-trained base model: 50GB (Llama 3 70B)
- Download training dataset: 200GB
- Export checkpoints after each experiment: 50GB × 5 experiments = 250GB
- Download final fine-tuned model: 50GB
- Total egress per training project: 550GB
- io.net cost: $0 (within free tier)
- AWS cost: $40 (450GB × $0.08/GB after 100GB free)
2. Distributed Training Across Regions:
If you train in US-West but analyze results in US-East or locally:
- Transfer checkpoints between regions: 100GB/week × 4 weeks = 400GB
- AWS charges $0.02/GB for inter-region transfer + $0.08/GB for final egress
- io.net treats inter-region within network as free (only final egress to internet counts)
3. Inference API Response Data:
Serving LLM inference APIs generates egress from response tokens:
- 1 million requests × 500 tokens/response × 2 bytes/token = 1GB
- 100 million requests/month = 100GB egress
- io.net: Free (within 1TB tier)
- AWS: $8/month
4. Debugging and Log Exports:
Exporting logs for analysis:
- TensorBoard logs: 5-10GB per training run
- Model debugging traces: 2-5GB per experiment
- Weekly exports: 50-100GB/month
- Often forgotten in cost estimates, adds $4-8/month on AWS
Ingress (Data Upload) Costs
Good news: io.net charges $0 for data ingress (uploading data to GPUs).
| Provider | Ingress Cost |
|---|---|
| io.net | Free |
| AWS | Free |
| Azure | Free |
| GCP | Free |
| CoreWeave | Free |
All major providers offer free ingress. Where io.net stands out is egress (downloads).
Regional Egress Pricing
Some providers charge different egress rates by region:
io.net:
- Same egress pricing regardless of region (1TB free, then $0.05/GB globally)
- No premium for APAC or South America
AWS (for comparison):
- North America/Europe: $0.08-$0.12/GB
- Asia Pacific: $0.08-$0.15/GB (20% premium)
- South America: $0.15-$0.25/GB (up to 3x premium)
Savings for Global Teams:
If your team is in Asia or South America, io.net's flat global pricing saves an additional 40-80% on egress vs. AWS regional premiums.
Egress Fee Optimization Strategies
How to minimize data transfer costs on any platform:
1. Use Persistent Storage Within the Cloud:
Instead of downloading data repeatedly, store it in cloud-attached volumes:
# Attach persistent volume to GPU
io launch --gpu A100 --storage 1TB --storage-type persistent
# Data persists between GPU sessions, avoiding re-download
Cost: $0.05/GB/month for storage vs. $0.05-$0.12/GB for repeated egress.
2. Compress Data Before Transfer:
Compress models and datasets before downloading:
# Compress model checkpoint before download
tar -czf model.tar.gz model_weights/
# 50GB model → 20GB compressed = 60% egress savings
3. Stream Large Files Instead of Downloading:
For analysis, stream data directly to local tools without full download:
# Stream TensorBoard logs without downloading
io logs stream --instance-id <id> | tensorboard --logdir -
4. Use Multi-Region Data Replication:
Replicate frequently accessed datasets across regions to avoid repeated cross-region transfers:
io storage replicate --source us-west --destination eu-west --path /datasets/
5. Leverage io.net's Free 1TB Tier Strategically:
If your team uses 800GB/month across multiple projects:
- Prioritize high-value transfers (final models, critical results)
- Keep intermediate checkpoints in cloud storage
- Delete unused data to stay under 1TB
Comparing Total Cost of Ownership (TCO)
Egress fees impact overall cloud costs significantly:
Scenario: Train and deploy Llama 3 8B model over 3 months
Workload:
- Download base model: 15GB × 3 iterations = 45GB
- Download training data: 100GB (one-time)
- Export fine-tuned models: 15GB × 10 experiments = 150GB
- Export experiment logs: 50GB
- Inference API responses: 200GB/month × 3 months = 600GB
- Total egress: 945GB
Cost Breakdown:
| Provider | GPU Cost (3 months) | Egress Cost | Storage Cost | Total TCO |
|---|---|---|---|---|
| io.net | $2,160 (A100 @ $1.20/hr × 60 hrs/month) | $0 (under 1TB free) | $75 (500GB @ $0.05/GB/mo × 3) | $2,235 |
| AWS | $5,508 (A100 @ $3.06/hr × 60 hrs/month) | $76 (845GB × $0.09/GB) | $150 (500GB @ $0.10/GB/mo × 3) | $5,734 |
| Azure | $5,220 (A100 @ $2.90/hr × 60 hrs/month) | $82 (945GB × $0.087/GB) | $135 (500GB @ $0.09/GB/mo × 3) | $5,437 |
io.net saves $3,202-$3,499 (56-61%) including egress fees.
When Egress Fees Matter Most
Workload types where egress costs significantly impact TCO:
High Egress Impact Workloads:
1. Model exporting for production deployment - Downloading 20-50GB models regularly
2. Dataset preparation workflows - Frequent uploads/downloads for preprocessing
3. Multi-cloud strategies - Syncing data between io.net and AWS/Azure for redundancy
4. Distributed teams - Multiple team members downloading results for analysis
5. Edge deployment - Transferring models to edge devices (robotics, IoT)
Low Egress Impact Workloads:
1. Inference APIs hosted entirely on io.net - Responses stay within network
2. Continuous training pipelines - Data stays in persistent storage
3. Batch processing - Results are small summaries, not raw data
Enterprise Egress Discounts
For teams transferring 50TB+/month:
io.net Enterprise Egress Tiers:
- 50-100TB/month: $0.04/GB (20% discount)
- 100-500TB/month: $0.03/GB (40% discount)
- 500TB+/month: Custom pricing (typically $0.02-$0.025/GB)
Contact [email protected] for custom egress pricing.
AWS Reserved Egress Pricing (for comparison):
AWS doesn't offer reserved egress discounts - you pay standard rates regardless of commitment. Only way to reduce egress is through AWS Direct Connect ($0.02/GB) which requires $500/month minimum plus hardware costs.
Egress Fees for Specific Use Cases
Use Case 1: Fine-Tuning Llama 3 70B
- Download base model: 140GB (one-time)
- Download training data: 500GB (one-time)
- Export fine-tuned model: 140GB
- Export evaluation results: 10GB
- Total egress: 790GB
- io.net cost: $0 (under 1TB free)
- AWS cost: $60 (690GB × $0.087/GB after 100GB free)
Use Case 2: Stable Diffusion Image Generation Service
- Download SD XL model: 7GB (one-time)
- Generated images served to users: 100,000 images/month × 5MB avg = 500GB/month
- Monthly egress: 507GB
- io.net cost: $0 (under 1TB free)
- AWS cost: $41 (407GB × $0.10/GB after 100GB free)
Use Case 3: High-Volume Inference API
- Serve 500M tokens/day (~1GB/day = 30GB/month response data)
- Minimal egress, falls well under free tier on any provider
- io.net cost: $0
- AWS cost: $0 (under 100GB free)
Use Case 4: Data Science Team (10 researchers)
Each researcher downloads:
- 50GB datasets/month
- 20GB experiment results/month
- Total: 700GB/month team-wide
- io.net cost: $0 (under 1TB free)
- AWS cost: $48 (600GB × $0.08/GB after 100GB free)
Monitoring Your Egress Usage
Track data transfer to avoid surprise bills:
io.net Dashboard:
- Real-time egress meter showing usage toward 1TB free tier
- Alerts when approaching 800GB (80% of free tier)
- Breakdown by instance and region
API Monitoring:
# Check current month's egress usage
curl -X GET https://api.io.net/v1/billing/egress \
-H "Authorization: Bearer <your-api-key>"
# Response:
{
"current_month": "2026-04",
"egress_gb": 673,
"free_tier_remaining_gb": 327,
"overage_cost": 0.00,
"projected_monthly_cost": 0.00
}
Cost Alerts:
# Set alert when egress exceeds 1TB (overage begins)
io alerts create --metric egress --threshold 1000 --action email
Related Questions
Does io.net charge for data transfer between GPUs in the same region?
No. Data transfer between instances within the same io.net region is completely free. This is important for distributed training where GPUs communicate frequently. Only data leaving the io.net network to the public internet counts as egress. Inter-GPU communication, storage access, and internal networking are all free, making multi-GPU clusters cost-effective.
How does the 1TB free egress tier work for multiple projects?
The 1TB free tier applies to your entire account across all projects and instances. If you have 5 projects, they collectively share the 1TB free allowance. To track usage by project, use io.net's tagging feature to monitor egress per project, then allocate the shared free tier accordingly. For teams needing separate billing, create multiple accounts (each gets 1TB free).
Are there egress fees for serving inference APIs to end users?
Yes, but they're minimal. Inference API responses count as egress, but LLM responses are small (1KB-10KB per request). Serving 1 million inference requests with 500-token responses (~1KB each) uses only 1GB of egress. Most inference workloads stay well under the 1TB free tier. For ultra-high-volume APIs (100M+ requests/month), egress is still 40-60% cheaper on io.net than AWS.
Can I avoid egress fees by keeping data in io.net storage?
Yes. Data stored in io.net's persistent volumes doesn't incur egress fees until you download it to external systems. For long-running projects, keep datasets, models, and checkpoints in persistent storage ($0.05/GB/month) and only download final results. This eliminates most egress costs while providing fast access for all your GPU instances.
What happens if I exceed 1TB egress in a month?
You're automatically charged $0.05/GB for usage beyond 1TB. There's no service interruption or throttling. For example, if you use 1.5TB egress in a month, you pay $0 for the first 1TB and $25 for the additional 500GB (500GB × $0.05/GB). Billing is transparent and appears on your monthly invoice. Set up alerts at 800GB to get notified before overage charges begin.
View complete pricing → or start deploying →
Last updated: April 2026 | Egress pricing comparison based on published rates from AWS, Azure, GCP as of April 2026
