Cloud Egress Cost in 2026: The Data Act, Cloudflare R2, and the End of the Tax
Egress fees are losing their grip. The European Data Act, Cloudflare R2, Hetzner economics, and the cross-AZ tricks that actually move the bill.
Cloud egress fees have been the single most loathed line item in enterprise cloud bills for a decade. The story changed sharply between 2024 and 2026. The European Data Act came into force, AWS and GCP made unprecedented egress-fee concessions for customers leaving their clouds, Cloudflare R2’s zero-egress model pushed object-store pricing toward a new equilibrium, and the hyperscalers quietly tightened cross-AZ and cross-region pricing where they could. The net effect: egress is no longer the cost trap it was, but the new shape of the bill demands a different strategy.
This is the 2026 platform-engineer view of where egress savings actually come from.
The European Data Act ban, in practical terms#
The EU Data Act took full force on 12 September 2025. Two provisions hit egress directly:
- Switching charges between cloud providers must be eliminated by January 2027. Through the transition period, providers are required to reduce switching charges to cost-recovery rates only.
- Functional equivalence and parallel-use must be supported. Providers cannot impose technical or commercial barriers that effectively trap data in.
The implementation pattern: AWS, Google Cloud, and Azure all announced free egress for customers fully migrating off their respective clouds in 2024 and 2025 (the announcement preceded the Act and was widely understood as positioning for it). The details differ — AWS requires you to be closing your account, GCP requires you to be migrating workloads off, Azure has its own paperwork — but the headline is real: leaving the cloud no longer costs a 7-figure egress bill at most scales.
What it does not change: ongoing egress for production traffic that stays on the cloud. The Data Act targeted lock-in, not normal egress, and the per-GB charges on production traffic remain.
What egress actually costs in 2026#
Rough late-Q1 2026 list pricing for egress to the public internet from US regions:
| Provider | Tier 1 (first 10 TB/month) | Tier 4 (50 to 150 TB) | Negotiated EDP (high volume) |
|---|---|---|---|
| AWS | 0.09 per GB | 0.05 per GB | 0.02 to 0.04 per GB |
| GCP | 0.12 per GB | 0.08 per GB | 0.02 to 0.04 per GB |
| Azure | 0.087 per GB | 0.05 per GB | 0.02 to 0.04 per GB |
| Cloudflare R2 to internet | 0 (zero egress) | 0 | 0 |
| Backblaze B2 to internet | 0.01 per GB (often free via Bandwidth Alliance) | 0.01 per GB | negotiated |
| Hetzner / OVH cloud bandwidth | usually included in instance cost | usually included | usually included |
The hyperscaler list rates are roughly half what they were in 2020 — a real concession the providers do not advertise. Negotiated EDP rates can pull the per-GB rate to 0.02 to 0.04 at large volume, which materially closes the gap with R2-style models for hot egress.
Cloudflare R2 and the zero-egress thesis#
Cloudflare R2 launched in 2022 as an S3-compatible object store with zero egress fees. By 2026 it is a real production option with documented adoption at major SaaS vendors (Shopify storefront images, Discord media, large publishing platforms, AI training-dataset sharing).
The economics: R2 storage at roughly 0.015 per GB-month is competitive with S3 standard. With zero egress, the total cost for read-heavy workloads is dramatically lower. For a dataset served 10x per month, R2 lands 60 to 80 percent below S3 + standard egress.
Where R2 wins: media serving, AI dataset distribution, static site assets, software distribution.
Where R2 is the wrong call: workloads where the rest of the AWS or GCP stack is the dominant integration cost, and the cross-cloud egress to get data into R2 erases the savings.
AWS Direct Connect and the private path#
For workloads with sustained large egress to on-prem, partner networks, or other clouds, AWS Direct Connect (and its equivalents — GCP Partner Interconnect, Azure ExpressRoute) is the right play. The economics:
- A 1 Gbps Direct Connect port is roughly 220 USD per month plus port-hour charges.
- A 10 Gbps Direct Connect port is roughly 1,640 USD per month plus port-hour charges.
- Data transfer over Direct Connect is roughly 0.02 to 0.03 per GB (versus 0.05 to 0.09 per GB over the public internet).
The breakeven at the 1 Gbps tier is roughly 5 to 10 TB per month of sustained egress. Below that, public internet egress is cheaper. Above 50 TB per month, the savings are substantial.
For SaaS clients that need to move data between AWS and partner networks routinely, we typically deploy a Direct Connect setup once egress crosses 20 TB monthly.
Hetzner, OVH, and the European economics#
Hetzner Cloud and OVHcloud have been the practical European answer to hyperscaler egress for years. The model: instance pricing includes a generous bandwidth allowance (Hetzner allocates 20 TB monthly for most instance types as of 2026), with overage at 1 EUR per TB — a fraction of hyperscaler rates.
For workloads that fit the Hetzner / OVH shape — predictable, not requiring the deep cloud-native ecosystem, fine with European regions — the egress economics are dramatically better. We see clients running production European workloads on Hetzner at roughly 30 to 50 percent total cost versus equivalent AWS deployment, with egress savings the dominant factor.
Where this falls apart: workloads tightly coupled to AWS- or GCP-native services (DynamoDB, Bedrock, Vertex AI). Hetzner does not have equivalents; the migration cost is real.
The cross-AZ vs cross-region pricing trick#
The egress-cost optimization most often missed: cross-availability-zone traffic inside a region is not free on AWS or GCP. It is billed at roughly 0.01 to 0.02 per GB in each direction. For a chatty microservices architecture spanning 3 AZs, this can be a real bill — sometimes larger than the egress to the public internet.
The patterns that reduce it:
- AZ-affinity routing. Use topology-aware routing in Kubernetes (Topology Aware Hints) so that pods prefer same-AZ destinations.
- Service mesh AZ-aware load balancing. Linkerd, Istio, and Cilium all support topology-aware load balancing that prefers same-AZ endpoints.
- Stateful service co-location. Place the application and its primary data store (Redis, Postgres replica) in the same AZ for the read path.
- Application-level batching. Reduce chattiness across AZ boundaries by batching small RPCs.
For a moderately chatty production workload, deploying topology-aware routing pulls cross-AZ data transfer by 40 to 70 percent.
Cross-region traffic is more expensive (0.02 to 0.09 per GB depending on cloud and region pair) and almost always avoidable through good architectural choices — replicate data into the region that consumes it, do not pull it across.
Multi-CDN strategies#
For media-heavy applications, the egress-cost optimization comes from offloading the bulk of read traffic to a CDN before it ever crosses the cloud egress meter. The 2026 patterns:
- Cloudflare in front of any cloud origin. Cloudflare’s cache-hit rate on well-tuned applications is 80 to 95 percent, and the bandwidth in is effectively free under most Cloudflare plans. Egress from AWS to Cloudflare is the same per-GB rate as public-internet egress, but it happens once per cache-eviction, not once per request.
- Multi-CDN with Cloudflare + Fastly or Cloudflare + CloudFront. For very high availability or global presence, multi-CDN routing via a DNS-based load balancer (NS1, Cedexis, Cloudflare Load Balancing) means cache hits in either CDN serve the user without further origin egress.
- Origin shield. A second-tier cache layer (Cloudflare’s origin shield, CloudFront’s origin shield) reduces origin egress further by deduplicating cache misses from multiple PoPs.
For a SaaS application serving images, videos, or large software downloads, a well-tuned CDN setup pulls cloud egress by 80 to 95 percent.
How we audit egress in 2026#
For client engagements, the workflow:
- Pull the full cost-by-usage-type export from AWS Cost and Usage Report, GCP Billing Export, or Azure Cost Management.
- Identify the top egress line items: public-internet egress, cross-AZ, cross-region, inter-account.
- Profile the source services: which workloads, which destinations.
- For each top line: apply the right pattern — CDN, AZ affinity, Direct Connect, R2 migration for object-heavy workloads, regional consolidation for cross-region traffic.
The savings are workload-dependent but routinely in the 30 to 60 percent range on the egress line for clients that have not previously focused here.
The thing the cloud will not optimize for you#
Cloud providers ship features that reduce egress (S3 Transfer Acceleration, AWS Global Accelerator, CloudFront PrivateLink, GCP Cloud CDN). They do not ship features that reduce your bill by suggesting you move to a competitor. The savings work above only happens if somebody owns it.
For the broader cost question, see our FinOps reserved instances piece, the FinOps cloud cost optimization take, and our Kubernetes production patterns for the architectural moves that make egress optimization easier.
Where pdpspectra fits#
Our cloud infrastructure and DevOps and CI/CD practices include egress-cost audits and remediation as a standard FinOps offering. We have pulled egress costs 40 to 70 percent for SaaS and media clients through CDN tuning, AZ-affinity routing, and object-store migration to R2 where appropriate.
Related reading: FinOps reserved instances, FinOps cloud cost optimization, and Kubernetes production patterns.
Egress costs respond to architecture, not begging the cloud rep for a discount. Talk to our team about your egress bill.