Cloud Data Egress Cost Analysis: Comparing 44 Providers

I Compared Data Egress Costs Across 44 Cloud Providers — Here’s the Breakdown

Faiz, the developer behind GPUPerHour.com, manually audited 44 cloud providers to benchmark data transfer costs. He found that while some providers offer unlimited free egress, others charge up to $0.55/GB. This disparity creates massive hidden costs for data-intensive workloads like machine learning.

Why This Matters

For machine learning practitioners, egress is not a secondary expense but a primary architectural constraint. Training datasets often exceed 100GB, and saving model checkpoints every few steps can generate terabytes of outbound traffic. While the GPU hourly rate is the most visible metric, a $90/TB egress fee on AWS can quickly eclipse compute savings, making provider selection critical for total cost of ownership. Selecting a provider like Lambda or RunPod with zero egress fees allows for frequent checkpointing without financial penalty. This reality often conflicts with ideal multi-cloud models that assume low friction for data mobility.

Key Insights

• 50% of GPU cloud providers, including RunPod, Lambda, and Vast.ai, offer $0 egress fees as of 2026.
• AWS and Google Cloud maintain high margins with egress rates of $90/TB and $120/TB respectively.
• Hetzner offers a high-value alternative with 20TB free and overages at $1.18/TB, which is 76x cheaper than AWS.
• Developer-focused platforms like Netlify and Render charge the highest premiums, with rates reaching $550/TB and $300/TB.
• Cloudflare R2 and OVHcloud provide zero-cost egress, serving as critical infrastructure for reducing storage-related data transfer bills.

Practical Applications

• Training LLMs on Lambda or RunPod: Utilizing providers with free egress prevents unexpected costs when saving 50GB model checkpoints to external storage.
• Avoiding the Netlify Bill Shock: Scaling high-traffic sites on developer platforms can lead to $100k+ invoices due to $550/TB overage rates compared to dedicated servers.
• Multi-cloud data shuffling: Using Hetzner or OVHcloud for data-heavy components reduces the egress tax incurred when moving terabytes between service regions.

References:

• https://dev.to/gpuperhour/i-compared-data-egress-costs-across-44-cloud-providers-heres-the-breakdown-4m5o
• https://gpuperhour.com/reference/data-egress