object-storage · Educational

Cloud egress: AWS S3 vs Google Cloud Storage path matrix

S3 and GCS Internet GETs, private compute pulls, replication, and CDN origin fill are different path classes. Align those templates before comparing any $/GB cell — replication bytes are not Internet GETs. Knit S3 CRR egress when bucket→bucket inter-region replication is the driver, and the AWS vs GCP spoke for broader multi-cloud path templates. Unknown until verified on dated provider pricing.

Updated 2026-09-16

Align S3 and GCS path classes — Internet GETs, private compute pulls, replication, and CDN origin fill — before comparing any $/GB cell.

Path matrix: AWS S3 vs Google Cloud Storage egress path classes (educational)
Path class	AWS S3 (typical)	GCS (typical)	Verify
Internet egress (public GET)	Tiered Internet DT-out after free grant	Network egress / download tiers	Region + free allotment
Same-region compute pull	VPC gateway/interface endpoint patterns	Private Google Access / PSC patterns	Private path fees
Cross-region / multi-region	CRR / MRC + inter-region DT	Multi-region / dual-region bucket meters	Replication ≠ user downloads
CDN origin fill	CloudFront / other CDN origin	Cloud CDN / Media CDN origin	Who bills miss bytes
Free-tier cliff	S3 / account DT-out grants	GCS always-free / Free Trial interactions	Model cliffs explicitly

Provider source footnotes

AWS S3 pricing — as-of 2026-09-09 Unknown until verified
GCP Cloud Storage pricing — as-of 2026-09-09 Unknown until verified
GCP network pricing — as-of 2026-09-09 Unknown until verified

Dated starting points only — Unknown until verified means no invented $/GB. PipeToll does not scrape private consoles or guarantee figures.

This spoke is a path matrix for object-storage egress on AWS S3 versus Google Cloud Storage. Titles lead with Cloud egress so readers are not hunting American pipe-toll brands. PipeToll refuses invented list prices — cells stay Unknown until verified.

Path templates to model

Client ← Internet ← S3/GCS — classic public GET; free grants and tiered Internet download meters apply after cliffs.
VPC/VPC-SC compute ← private path — gateway/interface endpoints or Private Google Access / PSC shift meters; watch private-path fees.
CRR / multi-region replication — bytes that never touch an end user still bill; keep separate from user downloads.
Client ← CDN ← object origin — miss-driven origin fill may sit on storage or CDN bills depending on pairing.
Analytics export S3 ↔ GCS — the source usually pays egress; dual-running months understate cutovers if ignored.

Common traps

One blog $/GB across clouds — Pasting a single “egress $/GB” from a blog across S3 and GCS ignores free-tier cliffs, Class A/B or GCS ops classes, and CDN origin bills. Align path classes first on this matrix and the AWS vs GCP spoke.
Multi-region = free HA — Treating multi-region / dual-region placement as “free HA” hides replication and placement meters that never touch an end user. Keep them off the Internet GET line.
CRR blended with public GET — Blending S3 CRR / GCS dual-region replication GB with public Internet GET GB in one cell invents a fake unit rate. Keep replication on its own Mode A line (S3 CRR).
Private endpoints zero every meter — Assuming VPC endpoints / Private Google Access / PSC erase every line item is unsafe. Private-path fees, residual Internet, and ops classes still stack — sibling VPC endpoints vs NAT.
CDN origin fill pasted into GET — Miss-driven origin fill is Mode C, not Internet GET and not CRR. Use the origin-fill estimator or multi-CDN spoke.
Scorecrowns from Incomplete — No scorecrowns from Incomplete cells — badge Unknown until verified. No guaranteed savings from switching logos.

FinOps checklist

Inventory top prefixes by GET volume and bytes on each cloud.
Separate Internet, private, replication, and CDN-origin lines; open S3 CRR when AWS replication rules drive the bytes.
Model free-tier cliffs with the AWS cliff and multi-cloud cliffs spokes.
Open calculator Mode A (?mode=gb&path=s3-gcs) for Internet GET sensitivity; use a second Mode A line (or ?path=s3-crr) for replication GB.
Cite dated provider pages per methodology; share vocabulary via the glossary and the AWS vs GCP path-template spoke.

FinOps-lite Unknown until verified

Worked example (educational): 6 TB public GET + 800 GB replication

Path A: clients download 6,000 GB from object storage over Internet (public GET).
Path B: 800 GB of cross-region / multi-region replication in the same month — not user downloads.
Run path templates separately in Mode A — do not blend replication with user egress. Paste dated $/GB or leave cells Unknown.
Optional Path C: 1,200 GB same-region private compute pull (VPC endpoint / Private Google Access) — private fees replace some Internet meters; do not paste Internet $/GB.

Numbered S3 vs GCS path-class stack — Unknown rate cells
#	Path class	Volume	Unit rate
1	Internet public GET (S3 DT-out / GCS download)	6,000 GB	Unknown $/GB — ?path=s3-gcs
2	CRR / multi-region replication	800 GB	Unknown $/GB — ?path=s3-crr
3	Private compute pull (VPC-EP / PGA)	1,200 GB	Unknown private-path fees
4	CDN origin fill (if any)	as measured	Unknown — Mode C
5	Ops / Class A-B / GCS ops classes	request counts	Often dominate — not $/GB alone

Second educational stack (private-path twin): same 1,200 GB compute→bucket pull — AWS same-region VPC endpoint / PrivateLink-style path vs GCP Private Google Access / PSC. Private fees can displace Internet DT-out; still keep replication (#2) and CDN origin fill (#4) on their own lines. Free-tier cliffs for Internet downloads sit beside #1 via AWS free-tier cliff and multi-cloud cliffs. Sibling multi-cloud templates: AWS vs GCP egress. Synthetic only — rates Unknown; no migration savings guarantee.

Worked-estimate sensitivity (educational)

Hold the 6 TB public GET + 800 GB replication baseline and walk three S3 / GCS path-class runs. Internet $/GB, replication $/GB, and private-path fees stay Unknown until dated provider cites — literacy arithmetic, not a quote or migration ROI.

Run A — Internet GET + CRR split: 6,000 GB public GET on Mode A (?path=s3-gcs) plus 800 GB replication on a second Mode A line (?path=s3-crr). Never blend into one cell.
Run B — CRR-only twin: same month with zero user downloads but 800 GB (or larger) bucket→bucket / multi-region replication still bills. Re-run CRR Mode A alone; sibling depth on S3 CRR egress.
Run C — Private-path sibling: 1,200 GB same-region compute→bucket pull (VPC endpoint / PGA / PSC) displaces some Internet DT-out — private fees replace Internet $/GB, not CRR. Free-tier cliffs for Internet leave sit beside Run A via AWS cliff and multi-cloud cliffs.

Sensitivity runs — Unknown rates (educational)
Run	What changes	Meters to keep visible	Unit rates
A	6 TB GET + 800 GB CRR split	Internet GET · replication (separate Mode A)	Unknown
B	CRR-only twin (0 user GET)	Replication / multi-region placement only	Unknown
C	1.2 TB private-path sibling	Private fees · residual Internet · cliffs beside A	Unknown

Cross-read S3 CRR, AWS vs GCP, S3 vs R2, CDN origin fill, and methodology. Badge every unverified rate Unknown until verified. No savings guarantee from switching object-store logos.

Open calculator · S3/GCS GET 6 TB Mode A · S3 CRR 800 GB Mode A · 1.2 TB private-path sibling S3 CRR egress AWS vs GCP

FAQ on this path

Is GCS egress the same SKU as S3 Internet DT-out?

No. Map identical path classes first (Internet, private, replication, CDN origin), then apply each provider’s dated rates.

Is replication the same meter as Internet GET?

No. Public Internet GETs are user-download egress; S3 CRR / GCS multi-region or dual-region replication moves bucket→bucket (or placement) bytes that never touch an end user. Keep separate Mode A lines — open S3 CRR egress when AWS replication rules are the driver, and use the AWS vs GCP spoke for cross-cloud path templates.

Do multi-region buckets remove user egress?

Multi-region / dual-region placement changes replication and availability meters — it does not erase Internet download billing. Keep lines separate.

Should I compare only storage $/GB?

Never. Request ops, free-tier cliffs, and CDN origin fill often dominate. See methodology.

Is a VPC endpoint / Private Google Access pull the same as Internet GET?

No. Private compute→bucket paths shift meters (endpoint hours, private GB, PGA/PSC) and can displace some Internet DT-out — still keep them off the public GET line. Deep-link ?path=s3-gcs only after naming the path class; sibling private-path literacy on VPC endpoints vs NAT.

Where does CDN origin fill sit in an S3 vs GCS compare?

Miss-driven fill is its own Mode C line — not Internet GET and not CRR. Use the origin-fill estimator or multi-CDN spoke.

Trademarks: Amazon Web Services, AWS, Microsoft Azure, Google Cloud Platform, GCP, and Cloudflare are trademarks of their respective owners. PipeToll (Ameti One LLC) is not affiliated with, endorsed by, or sponsored by those companies. Educational comparisons may be outdated — verify current rates with each provider.