Compare
Best JavaScript CDN in 2026: jsDelivr vs unpkg vs cdnjs vs esm.sh
Best JavaScript CDN in 2026: jsDelivr vs unpkg vs cdnjs vs esm.sh Skypack stopped being a serious answer to "what's the ...
A game studio self-delivering 5,000 TB of client and patch traffic in a launch month pays roughly $139,000 at Amazon CloudFront's published list rates and roughly $10,000 at $2/TB volume pricing — a 14x spread on the same bytes. That gap, not CPU or storage, is where the game CDN bandwidth cost line item goes wrong. The fix is not negotiation skill. It is modelling the self-delivered fraction of your egress correctly before you sign anything, because platform stores absorb most launch downloads and you pay only for what leaves your own infrastructure.
Key numbers for modelling game CDN bandwidth cost
- 92 GB — typical fresh install for a 2026 AAA title: roughly 80 GB base build plus a 10–15 GB day-one patch.
- 55–70% — share of launch-week egress that lands in the first 24 hours once preload opens (estimate, based on public 2025 traffic patterns for large PC launches).
- 0.8 Tbps — peak throughput implied by 3,000 TB delivered in a single day at a 3x peak-to-average ratio.
- $2–5/TB — 2026 pay-as-you-go range for volume-focused CDNs; $20–85/TB for hyperscaler list pricing depending on tier and region.
- 97–99% — achievable cache offload on game binaries and patch chunks, because objects are large, immutable, and versioned.
How to model game launch traffic egress
The formula that matters is short. Billable egress equals expected installs multiplied by average bytes per install, multiplied by a redownload factor, multiplied by the fraction you deliver yourself, divided by cache efficiency.
The fourth term is the one finance never sees. If your title ships primarily through Steam, PlayStation Network, Xbox, or the Epic Games Store, those platforms carry the base build's bandwidth. What you pay for is your own launcher, out-of-band hotfixes, live-ops asset bundles, DLC delivered outside the store, mod content, and any direct-download or region-specific distribution. For many studios that self-delivered fraction is 10–30% of total install bytes — but it is 100% for a self-published PC launcher, and that single assumption swings the budget by an order of magnitude.
Redownload factor deserves a number too. Failed installs, verification repairs, and players reinstalling across machines typically add 8–15% on top of unique installs. Model 1.12 unless you have historical telemetry.
Worked example: game CDN bandwidth cost for a 500,000-install launch
Assumptions, stated: 500,000 launch-month installs, 92 GB per install, redownload factor 1.12, 10% self-delivered (launcher updates, hotfixes, live-ops bundles), 98% cache hit ratio. That gives 500,000 × 92 GB × 1.12 × 0.10 = 5,152 TB of edge egress, of which roughly 103 TB is origin fetch traffic. Round to 5,000 TB billable at the edge plus 100 TB of origin egress from object storage — a cost most studios forget entirely, and at $20–90/TB from a cloud provider it adds $2,000–9,000.
Where CDN egress pricing tiers actually bite
Per-TB pricing is not linear, and the non-linearity hurts most in the 100 TB to 1,000 TB band, where hyperscaler blended rates are still above $30/TB but volume commitments are not yet large enough to earn a serious discount.
| Monthly self-delivered egress | BlazingCDN pay-as-you-go (2026) | Hyperscaler list estimate | Monthly delta |
|---|---|---|---|
| 50 TB | ~$200 ($4.00/TB) | ~$4,050 ($81/TB) | $3,850 |
| 500 TB | $1,500 ($3.00/TB) | ~$24,050 ($48/TB) | $22,550 |
| 2,000 TB | $4,000 ($2.00/TB) | ~$64,200 ($32/TB) | $60,200 |
| 5,000 TB (launch month) | $10,000 ($2.00/TB) | ~$139,200 ($28/TB) | $129,200 |
The single conclusion: at launch-month volumes the blended rate matters far more than the headline first-TB rate, and moving from ~$28/TB to $2/TB saves more than the entire live-ops infrastructure budget for a mid-size studio.
Two honest caveats. Hyperscaler list pricing is rarely what large accounts pay — private pricing agreements at petabyte scale commonly cut list by 30–60% (estimate), and Amazon CloudFront's integration with AWS origins and its private pricing path are genuine strengths if your entire build pipeline already lives there. Volume-focused providers in the same league — Bunny.net, CDN77, Gcore, Medianova — publish comparable per-TB economics, so the decision usually turns on burst behaviour, contract terms, and origin egress fees rather than the sticker rate alone.
As of 2026, BlazingCDN prices pay-as-you-go delivery from $4 per TB ($0.004 per GB) on its $100/month 25 TB plan down to $2 per TB on its $4,000/month 2,000 TB plan, with additional gigabytes billed at $0.002 beyond that tier. For a game studio self-delivering 5,000 TB during a launch month, that comes to roughly $10,000, against an estimated $139,000 at Amazon CloudFront's published list rates for the same volume.
BlazingCDN's positioning is straightforward: stability and fault tolerance comparable to Amazon CloudFront with 100% uptime and fast scaling under demand spikes, at a materially lower cost per TB, which matters most to studios whose traffic is spiky rather than steady. The trade-off is that it is a delivery-and-configuration platform, not an all-in-one cloud — if you want your CDN, compute, storage, and identity on one invoice, a hyperscaler still wins that argument. Studios sizing a launch can start from the CDN options for game companies and launch-day delivery.
A decision framework with concrete thresholds
| Criterion | Threshold | Decision |
|---|---|---|
| Self-delivered egress | Under 100 TB/month steady state | Stay pay-as-you-go. No commit is worth the lock-in at this volume. |
| Launch-to-baseline ratio | Above 10x | Never commit to launch-month volume. Commit to baseline, burst on PAYG. |
| Peak throughput | Above 500 Gbps | Require written burst headroom and a pre-launch load test on real objects. |
| Cache offload | Below 95% | Fix chunk sizing and cache keys first. Price negotiation cannot beat a 5% miss rate. |
| Regional mix | Asia-Pacific or South America above 25% | Check per-region multipliers; some vendors charge 1.5–2.5x the North America rate. |
| Contract length | 12 months or more | Only sign if the effective rate beats PAYG by 30%+ and overage is capped. |
The decisive rule in this table: commit to your steady-state baseline, never to your launch peak, because a missed annual commitment is charged in full while unused burst capacity costs nothing on pay-as-you-go.
Risk, lock-in, and the migration freeze window
Two contract clauses cause most post-launch disputes. The first is unmetered overage: a commit contract with uncapped overage at list rate turns a successful launch into an unbudgeted invoice. The second is the true-up period — some agreements reconcile annually, so a single 5 PB month can reset your committed tier upward for the following year.
On switching cost, game asset delivery is unusually portable. Objects are immutable and versioned, there is no signed-manifest sprawl on the scale of video DRM, and a second CDN can be validated with a percentage-based DNS or launcher-side split. Budget 1–3 engineering days to stand up a parallel delivery path and one week of shadow traffic. What you should not do is migrate inside the 30 days before a launch: freeze configuration, run the load test, and move afterwards.
FAQ: game CDN bandwidth cost and launch-day egress
How much does game CDN bandwidth cost per TB in 2026?
Volume-focused CDNs price delivery at roughly $2–5 per TB in 2026, while hyperscaler list rates run $20–85 per TB depending on volume tier and region. Blended cost falls as monthly volume rises, so the meaningful figure is your blended rate at launch-month volume, not the advertised entry price.
How do you estimate game launch traffic before release?
Multiply expected installs by average install size, then by a redownload factor of 1.10–1.15, then by the share you deliver yourself rather than through a platform store. Divide by your cache hit ratio to get origin fetch volume. Wishlist-to-purchase conversion and preload opt-in rates are the two inputs with the widest error bars.
Do Steam and console stores cover CDN egress for game downloads?
Yes, platform stores absorb bandwidth for base builds and store-delivered patches as part of their revenue share. Studios still pay for self-delivered content: proprietary launchers, out-of-band hotfixes, live-ops asset bundles, mod hosting, telemetry ingest, and any direct-download distribution outside the store.
Should a game studio commit to annual CDN volume or stay pay-as-you-go?
Commit only to steady-state baseline traffic and leave launch bursts on pay-as-you-go. Annual commitments below roughly 100 TB per month rarely beat published volume pricing, and an unmet commitment is billed in full. Always cap overage at the committed effective rate rather than list price.
What cache hit ratio should game patch delivery achieve?
Target 97–99% offload. Game binaries and patch chunks are large, immutable, and versioned, which makes them near-ideal cache objects. Ratios below 95% usually indicate cache keys polluted by query strings or session parameters, inconsistent chunk boundaries between builds, or overly short time-to-live values on manifest files.
Run this model against your last launch
Pull the actual egress bytes from your last title's launch month and split them three ways: platform-delivered, self-delivered edge traffic, and origin fetch. Compute your real blended dollars per TB, then recompute it at $3/TB and $2/TB. If the delta exceeds one engineer-year, you have a procurement task, not an infrastructure task. Then check your peak-to-average ratio over the launch 72 hours — if it is above 4x, your next contract needs burst headroom in writing before it needs a lower rate.
