On May 1, Google Cloud doubled its CDN Interconnect egress rates across North America, Europe, and Asia. The North America rate jumped from $0.04 per gigabyte to $0.08. Europe climbed from $0.05 to $0.08. Asia moved from $0.06 to $0.085. Google cited “significant investments in global infrastructure” as the rationale and offered no advance negotiation window.
If you run a GCS bucket and serve content through Cloudflare, Akamai, or Fastly, your bill just went up automatically. CDN Interconnect pricing applies when traffic flows through allowlisted CDN providers, and there is no opt-out within GCP.
That change alone adds roughly $14,400 per year for a team pushing 30TB monthly through a CDN. But the bigger story is one FinOps programs keep ignoring: data transfer costs are the most under-optimized line item on enterprise cloud bills, and every structural trend in 2026 is making them worse.
The Egress Gap Across All Three Hyperscalers
Google gets the headlines for the CDN rate increase, but egress pricing across all three major providers remains one of the least competitive cost categories in cloud infrastructure.
AWS charges $0.09 per gigabyte for the first 10TB of internet egress monthly, stepping down at higher volumes. Cross-region transfer within AWS runs $0.01 to $0.02 per gigabyte. AWS also charges for inter-AZ traffic within a single region at $0.01 per gigabyte each direction. Teams running distributed microservices across availability zones often discover this cost only after their first detailed bill review. The AWS cost optimization playbook rarely starts with data transfer, but arguably should.
Azure starts at $0.087 per gigabyte for the first 5TB, roughly matching AWS. Inbound data transfer is free, but outbound tiers are complex and vary by region. Azure’s cost management tooling surfaces compute and storage waste effectively; data transfer visibility remains a gap.
GCP charges $0.12 per gigabyte at the Premium Tier (the default) for the first terabyte of internet egress. That is 33% more expensive than AWS at equivalent volumes, despite GCP’s lower base storage rate of $0.020 per gigabyte versus S3’s $0.023. The storage discount evaporates once data moves.
Oracle Cloud (OCI) breaks the pattern entirely: 10TB per month of free egress, then $0.0085 per gigabyte beyond that. Roughly 10x cheaper than AWS at equivalent scale.
The numbers crystallize when you model a real scenario. Sending 10TB of data to the internet costs $0 on Cloudflare R2, approximately $870 on AWS S3, and approximately $1,127 on GCP at Premium Tier rates. Same data, same outcome, wildly different cost.
Why Data Transfer Resists Standard FinOps Optimization
Most FinOps programs focus optimization energy on compute rightsizing and commitment management. Data transfer stays in the blind spot for structural reasons that have nothing to do with practitioner skill.
Egress is invisible at provisioning time. When an engineering team sizes a VM or provisions a storage bucket, compute and storage costs appear in pricing calculators. Egress lives in footnotes. Architectural decisions about multi-region redundancy and cross-cloud pipelines happen without modeling the data movement cost.
Resource tagging breaks down for network traffic. FinOps cost allocation relies on tagging. Network traffic between services, regions, and the internet resists attribution to a specific team or workload. It accumulates in shared or unallocated buckets that nobody owns and nobody optimizes.
Volume follows usage, not provisioning. Compute costs scale with capacity you provision. Egress scales with actual traffic patterns, API call volumes, and data pipeline throughput. A single integration pulling data across a cloud boundary every hour can generate thousands of dollars in monthly transfer charges without triggering standard anomaly thresholds.
Ingress is free by design. Every hyperscaler charges $0 for inbound data transfer. That deliberate asymmetry creates economic lock-in: getting data into the cloud costs nothing; getting it out is where the meter runs. The EU Data Act forced limited egress reductions in 2025, but only for customers fully exiting a cloud provider, not for ongoing operations.
During a fractional COO engagement through Ops Harmony, I worked with a media company that had optimized compute and storage aggressively but never examined their data transfer line. Egress was 18% of their total cloud bill. A single data pipeline replicating 2TB daily to a staging environment in a different region cost $14,000 annually, serving a staging environment that three engineers used occasionally.
AI Workloads and Multi-Cloud Are Multiplying the Problem
Two converging forces are accelerating egress costs in 2026, and neither is slowing down.
AI workloads generate massive data movement. Training requires pulling terabytes from storage to GPU instances, frequently across regions or availability zones because GPU capacity is scarce. A 10TB training dataset pulled across a provider boundary weekly produces roughly 40TB of billable transfer per month, costing approximately $4,000 in egress fees alone. The Flexera 2026 State of the Cloud Report found that cloud waste increased to 29% of total spend, the first rise in five years. AI data movement costs are a significant and largely unmeasured contributor.
Multi-cloud is the default architecture. Flexera reports that 87% of enterprises now operate multi-cloud environments. Every data flow crossing a provider boundary incurs egress from the source. Teams running inference on one cloud with data stored on another pay data transfer premiums that frequently exceed the compute costs themselves. For organizations refining their multi-cloud cost management approach, egress should be a primary variable in architecture decisions, not an afterthought discovered during quarterly bill review.
The combination is a multiplier: more data, moving more frequently, across more provider boundaries.
Five Approaches That Actually Reduce the Bill
Egress optimization is not glamorous, but the returns compound quickly.
Map your data flows before optimizing anything. Most teams cannot answer “how much data moves between which endpoints monthly?” Run a 30-day analysis using VPC Flow Logs (AWS), VNet flow logs (Azure), or VPC Flow Logs (GCP) to quantify the flows. Without that baseline, optimization decisions are guesswork.
Colocate compute with data. The cheapest egress is the transfer that never happens. If a workload reads heavily from a specific storage location, compute should run in the same region and ideally the same availability zone. GPU capacity constraints regularly push AI teams into cross-region architectures without accounting for the data movement premium. For organizations evaluating where workloads should run, data gravity deserves equal weight alongside raw compute cost.
Use private connectivity for high-volume predictable flows. AWS Direct Connect, Azure ExpressRoute, and Google Cloud Interconnect all offer lower per-gigabyte rates for consistent traffic. Teams moving 5TB or more monthly on a predictable schedule typically see net savings even after port and cross-connect fees.
Maximize cache-hit ratios at the edge. CDN caching reduces origin egress by serving repeat requests from edge locations. A well-tuned CDN configuration can reduce origin pulls by 70% or more. With Google’s CDN Interconnect rates doubling, the return on aggressive cache optimization just improved meaningfully: every request served from edge instead of origin now saves twice as much on peered egress.
Evaluate zero-egress alternatives for the right workloads. Cloudflare R2 ($0 egress), Backblaze B2 ($0 egress through the Cloudflare Bandwidth Alliance), and OCI (10TB free, then $0.0085 per gigabyte) offer dramatically lower data transfer costs. Archival data, backup storage, and static asset hosting are natural migration candidates where sub-millisecond latency from a specific hyperscaler is not required.
The Pricing Signal Worth Reading
Cloud prices are rising in 2026 across multiple categories. Google’s CDN egress rate doubling fits a broader pattern: compute and storage pricing competition tightens while data transfer pricing stays high because it creates switching costs that benefit the provider.
The practical FinOps response is to add egress to the optimization portfolio alongside compute rightsizing and commitment management. For any organization spending more than $10,000 monthly on cloud, a focused data transfer audit will almost certainly surface savings in the range of 5 to 15% of total spend. For AI and multi-cloud architectures, the number runs higher.
The cloud cost category that gets the least attention from FinOps teams is consistently the one with the most room to improve. Google just made that gap wider.