Google Cloud charges $0.08 to $0.12 per GB for standard internet egress and $0.01 to $0.05 per GB for inter-region traffic, and a large enterprise can cut those published rates by 30 to 80 percent inside an enterprise agreement. Egress is the line item buyers understand least and pay most for, because it scales with traffic rather than with provisioned capacity, and it rarely appears in the architecture diagram that gets the project approved. The buyers who control egress are the ones who price it before they build and negotiate it before they sign.
What egress is and why it dominates the bill
Egress is the charge for data leaving a Google Cloud region, whether it goes to the public internet, to a different Google Cloud region, or between zones inside a region. Ingress, data coming in, is free. That asymmetry is deliberate: it is cheap to move data into the platform and expensive to move it out, which is the commercial shape of every major cloud and the structural reason data has gravity. A workload that looks inexpensive on compute and storage can carry an egress bill several times larger than both, because every API response, every backup copied to another region, every analytics export, and every user download is metered on the way out. Egress is not a rounding error on a large estate; it is frequently the single largest variable cost and the one least visible in planning.
The Google Cloud egress rate card
Published egress pricing is tiered by destination and by volume. The rates below reflect Google Cloud general network pricing as of early 2026 and are the reference point a negotiation starts from, not the price a large buyer should accept.
| Egress type | Destination | Published rate (per GB) |
|---|---|---|
| Internet egress (standard tier) | Public internet | $0.085 to $0.12 |
| Internet egress (premium tier) | Public internet, first 1 TB lower band | $0.08 to $0.12 |
| Inter-region egress | Region to region, same continent | $0.01 to $0.02 |
| Inter-region egress | Cross-continent | $0.05 to $0.08 |
| Inter-zone egress | Zone to zone within a region | $0.01 |
| Cloud CDN cache egress | To internet via CDN | $0.02 to $0.08 |
| Egress on full provider exit | Leaving Google Cloud entirely | $0.00 (switching credit) |
Two details change the math. First, egress within the same region between most managed services and Compute Engine is free, so co-locating chatty components removes cost entirely. Second, since the European Data Act took effect, Google offers free egress to customers who fully exit the platform, which removes the historic exit penalty but does nothing for the ongoing operational egress that an active workload generates every day. The pricing structure these rates sit inside is covered in our Google Cloud licensing guide.
Where egress hides in an architecture
The egress that surprises buyers is almost never the obvious user-facing download. It is the architecture that crosses a billing boundary without anyone noticing. Cross-region replication of databases and object storage generates inter-region egress on every write. Multi-region high availability designs copy data continuously between regions for resilience and pay egress for the privilege. Analytics pipelines that export data out of BigQuery to an external tool, or back to an on-premise warehouse, meter on the way out. Backup and disaster recovery copies sent to a second region accumulate egress that is invisible until the monthly bill arrives. And any design that places a chatty microservice in one region and its dependency in another pays egress on every internal call. The first optimization is architectural, not commercial: keep traffic inside a region wherever the resilience requirement allows it.
The negotiation lever buyers miss: Egress is almost never on the standard discount schedule, so account teams treat it as fixed. It is not. A buyer with material, predictable egress can negotiate a custom egress rate, a waiver on inter-region replication traffic, or egress credits as part of a committed-spend agreement. The figure that unlocks the concession is your forecast egress volume, modeled by destination, presented as committed spend the vendor wants to capture. Bring that number to the table and egress becomes negotiable like everything else.
Committed-use and tiered discounts on egress
Google's headline committed-use discounts apply to compute and a defined set of services, and egress sits outside the standard committed-use program. That does not mean it is undiscountable; it means the discount comes through the enterprise agreement rather than through a self-service commitment. For a large estate, egress concessions are negotiated as part of the overall committed-spend deal: a custom price per GB, a volume tier that resets the rate above a threshold, or a pool of network credits that offsets the first defined volume each month. The mechanics of that agreement, and what is negotiable inside it, are set out in our GCP enterprise agreement guide, and the committed-spend discipline that governs it mirrors the approach in our GCP committed use discounts guide.
Free egress on exit, and why it changes little day to day
Regulatory pressure in Europe and competitive pressure across the industry pushed Google, like its rivals, to waive egress fees for customers who fully leave the platform. This removes the lock-in penalty that historically made switching prohibitively expensive, and it is a genuine improvement for buyers weighing an exit. It is also frequently misread. The waiver applies to the one-time bulk transfer when you leave; it does not touch the routine operational egress that a running workload generates every day to serve users, replicate data, and feed analytics. Treat the exit waiver as portability insurance, not as a reduction in your operating bill, and keep negotiating the operational rate.
The cross-cloud view
Egress pricing rhymes across the major clouds but differs enough to matter when you place a workload. AWS and Azure both charge for data transfer out on a tiered model, and the specific rates, the free allowances, and the inter-region treatment vary by provider and by region pair. A workload that moves large volumes to the internet, or that replicates heavily across regions, can have a materially different total cost on each platform once egress is counted, which is why egress belongs in the placement decision and not just the optimization backlog. Our cloud egress negotiation guide sets the providers side by side, and the AWS-specific fight is covered in AWS egress negotiation.
How to model egress before you commit
The discipline that controls egress is to forecast it by destination before the architecture is fixed. Separate the volume into three buckets: internet egress to end users, inter-region traffic driven by replication and resilience, and analytics or backup export leaving the platform. Each bucket has a different rate and a different optimization. Internet egress responds to caching and to a content delivery network that serves repeat traffic from cache rather than from origin. Inter-region traffic responds to architecture, keeping replication inside a region where the availability target permits. Export traffic responds to keeping analytics in place rather than shipping data out. Model the three, total them, and that number becomes both the optimization target and the committed-spend figure you negotiate against. The vendor hub for the full Google Cloud commercial position sits at Google Cloud advisory.
Governing egress after you sign
Egress is a running cost that drifts, so it needs monitoring the way any variable spend does. Tag traffic to the workloads and teams that generate it, alert when a service crosses an egress threshold, and review the top egress sources on a monthly cadence, because a single misconfigured replication job or a new analytics export can double the line item quietly. A negotiated egress rate also has a term, so the rate and any credit pool belong on the same renewal calendar as the rest of the agreement. Buyers who treat egress as a fixed cost of doing business pay the published rate forever; buyers who treat it as a managed, negotiated line cut it and keep it cut.
Cutting internet egress with a content delivery network
The largest single reduction available on user-facing internet egress comes from serving repeat traffic out of cache rather than from origin. Cloud CDN holds frequently requested objects at edge locations and serves them from there, so a popular file, image, or video is paid for once on the way into cache and then served many times at the lower cache-egress rate rather than at the full origin-egress rate each time. For any workload with a high ratio of repeat downloads, static assets, software distribution, media streaming, the cache hit rate is the variable that decides the bill, and a high hit rate can cut internet egress by more than half. The optimization is to maximize what is cacheable, set sensible cache lifetimes, and measure the hit rate as a first-class cost metric rather than a performance afterthought.
Standard versus premium network tier
Google offers two network tiers, and the choice changes both cost and performance. The premium tier routes traffic across Google's own global backbone for most of the journey, delivering lower latency and higher reliability at a higher egress rate. The standard tier hands traffic off to the public internet sooner, costing less but offering less control over the path and the performance. For latency-sensitive, user-facing traffic the premium tier usually earns its rate, while for bulk, machine-to-machine, or backup traffic where latency does not matter the standard tier can cut the cost materially. The discipline is to match the tier to the traffic rather than defaulting the whole estate to one tier, because a blanket premium-tier choice pays for performance that much of the traffic does not need. This tier decision sits inside the broader commercial picture covered in our GCP enterprise agreement guide.
A worked egress reduction example
The levers compound, and a single workload shows how much. Take an application serving 200 TB of internet egress a month at the standard published rate, with heavy cross-region replication adding another 80 TB of inter-region traffic for resilience. Three moves change the bill. Putting a content delivery network in front of the user traffic, with a 70 percent cache hit rate, shifts most of the 200 TB to the lower cache-egress rate. Confirming that the resilience requirement permits single-region replication for a portion of the workload removes a slice of the 80 TB of inter-region traffic entirely. And bringing the remaining forecast volume to the negotiation as committed spend earns a custom rate on what is left. Across the three, a buyer can take an egress line that looked fixed and cut it by a large fraction, which is the difference between accepting the published rate and engineering plus negotiating against it.
The buyer's takeaway
Egress is the most negotiable cost that most buyers never negotiate. Price it by destination before you build, keep traffic inside a region wherever resilience allows, bring your forecast egress volume to the table as committed spend, and negotiate a custom rate or a credit pool rather than accepting the rack rate as fixed. The exit waiver protects portability but does not lower your operating bill, so keep the operational rate on the negotiation list. We model GCP egress and negotiate the network terms through our cloud contract negotiation and software licensing advisory practices. The cheapest egress is the byte that never crosses a billing boundary; the next cheapest is the byte you negotiated a rate for in advance.