CDN for Video: The True Cost, the Best Options, and How to Cut Your Egress Bill

Start with the only number that matters, the one per viewer. A 1080p stream at 6 Mbps hands each viewer 5.4 GB over a two-hour event. A 4K stream at 25 Mbps hands them 22.5 GB. Every cost question in video delivery is that figure multiplied by your audience.

Run a two-hour 1080p event for 5,000 concurrent viewers and you have moved 27 TB. Do it weekly and you are at roughly 105 TB/month, which on AWS CloudFront's North America ladder bills at about $7,555/month — just over $90,000/year for one recurring show. Deliver the same bytes into Asia Pacific instead and the rate is 41% higher before you have changed a single line of config.

CDN for video isn't a utility cost. It's a tax on growth, and it scales linearly with audience size while your revenue per viewer does not.

This article shows exactly what you're being charged and why tiered pricing obscures the true number until the invoice arrives. It models TCO across providers at real traffic volumes, covers the technical mistakes that inflate egress unnecessarily, and gives you a step-by-step playbook for migrating off an expensive incumbent without breaking playback quality for a single viewer.

Why CDN Egress Pricing Is Designed to Hurt You at Scale

How Tiered Pricing Works Against High-Traffic Events

CloudFront's North America ladder runs $0.085/GB for the first 10 TB, $0.080 to 50 TB, $0.060 to 150 TB, $0.040 to 500 TB, then $0.030. (Rates: AWS CloudFront Price List API, July 2025) That descent looks meaningful on paper. In practice, most streaming teams never live in the cheap tiers.

Volume accumulates within a calendar month, then resets to zero. That reset is where the money goes: every month you pay the top $0.085/GB rate on the first 10 TB and $0.080/GB on the next 40 TB again, regardless of how much you delivered last month. The deepest published tiers — $0.030/GB and below — require sustained volume above 500 TB/month that most streaming teams never reach.

The math incumbents don't show you: 105 TB/month does reach the $0.060/GB tier — but only for the traffic above 50 TB. The blended rate still lands at $0.070/GB, because the expensive first 50 TB is re-billed at full price every single month. You never escape the top of the ladder; you only ever add cheaper traffic underneath it.

Regional Surcharges: Why APAC, LATAM, and Africa Cost 1.5–3x More

CloudFront charges $0.120/GB across most of Asia Pacific and $0.110/GB in South America on the first 10 TB — 41% and 29% above the North America baseline. Japan and Australia sit at $0.114/GB, India at $0.109/GB. (Rates: AWS CloudFront Price List API, July 2025) Fastly is steeper still outside North America: $0.19/GB for Asia, South America and Australia, and $0.28/GB for India, South Korea and Africa, against $0.12/GB in the US. (Source: Fastly published pricing) Akamai is enterprise-negotiated and unavailable without a sales engagement, so you cannot model your true cost until you're already deep in a commitment conversation.

For a team serving a global audience, regional surcharges aren't an edge case — they're a structural feature of how incumbents monetize non-NA traffic. Each region also runs its own independent tier ladder, so splitting an audience across three continents means climbing three separate ladders from the top rate rather than one.

The Overage Trap: How Predictable Streams Produce Unpredictable Invoices

Commit-and-overage contracts are the default enterprise offering from both Akamai and, increasingly, AWS at scale. They look like a path to lower unit pricing. They are actually a mechanism that transfers financial risk entirely to the buyer.

The clause that matters is the overage rate — typically priced at or above list rate once the committed volume is exceeded. A viral moment, a playoff game, a product launch that lands bigger than projected: all of these push you into overage territory at the worst possible time. The invoice arrives 2–4 weeks after the event. The spike is over. The damage is already done.

---

CDN for Video Cost Comparison: Real Per-GB Egress Rates Across Providers

Read the table below as total cost of ownership, not as a rate card. Published per-GB rates only tell you what a provider charges in its cheapest region at its most favourable tier. What determines your invoice is the blend: your regional traffic split, how bursty your events are, and whether the tier resets monthly.

North America and Europe: The Baseline Rates

All rates below are the published first-10-TB price in each region.

Notes: CloudFront rates are from the AWS Price List API (July 2025) and include 1 TB/month free. CloudFront steps down to $0.080/GB above 10 TB, $0.060 above 50 TB, $0.040 above 150 TB and $0.030 above 500 TB — per region, reset monthly. Fastly rates are from its published pricing page and include 100 GB/month free. Cloudflare's free CDN tier cannot legally carry your video: its Service-Specific Terms require Stream, Images or the Developer Platform for video delivery on non-Enterprise plans, and Stream bills $5 per 1,000 minutes stored plus $1 per 1,000 minutes delivered. Akamai publishes no rates. ZeroBuffer's rate is identical in every region but does tier down by volume: $0.0049 to 200 TB, $0.0045 to 500 TB, $0.0040 to 750 TB, $0.0035 to 1 PB.

APAC, LATAM, and Africa: Where Bills Actually Explode

The gap between NA and APAC/LATAM rates is a deliberate structural choice, not a reflection of infrastructure cost. Take the 27 TB event from earlier. Delivered entirely in North America it costs $2,211. Now move 40% of the audience to Asia Pacific — same bytes, same encode, same event:

• NA portion (60% = 16,200 GB): 10,240 × $0.085 + 5,960 × $0.080 = $1,347
• APAC portion (40% = 10,800 GB): 10,240 × $0.120 + 560 × $0.100 = $1,285
• Total: $2,632 — a 19% increase over the NA-only figure, from nothing but geography

The penalty compounds because each region restarts its own ladder at the most expensive tier. Splitting 27 TB across two regions means paying the first-10-TB rate twice.

Scale that to a global platform with 50%+ non-NA traffic and the surcharge compounding becomes the dominant cost driver. Africa rates, where published, run even higher.

Flat-Rate vs. Tiered: TCO Modeling at Three Traffic Levels

Every figure below is computed by walking each provider's published tier ladder, US/NA delivery, 1 TB = 1,024 GB. No blended shortcuts and no free-tier credits applied.

Note that the saving narrows as volume grows — CloudFront's blended rate falls from $0.085 to $0.048/GB between 10 TB and 500 TB, and Fastly's halves above its 10 TB step. Anyone quoting you a flat "94% cheaper" figure at every scale is not walking the tier ladder.

At 100 TB/month — roughly four weekly two-hour 1080p events to 5,000 concurrent viewers — the gap between CloudFront and $0.0049/GB is about $6,700/month, or $80,000/year. That is not a rounding error.

---

What Actually Kills Cache Hit Ratio on Video CDNs (And How to Fix It)

Cache Key Fragmentation: The Query Parameter Problem

Many video streaming CDN stacks create distinct cache entries for identical HLS segments because query parameters — tracking IDs, session tokens, analytics tags — are included in the cache key. A segment URL like seg-003.ts?uid=abc123&sid=xyz789 and seg-003.ts?uid=def456&sid=pqr012 are the same bytes but treated as separate cache objects. (Source: OTTVerse)

The result is a cache hit ratio that craters, origin fetches that multiply, and egress costs that increase in direct proportion to your concurrent viewer count — exactly when you can least afford it.

The fix is cache key normalization at the edge: strip all tracking and session parameters before the cache lookup, pass them downstream as headers if analytics systems require them, but never let them fragment the cache keyspace. Most CDN edge rulesets support this via header manipulation or query string ignore lists.

TTL Strategy for HLS/DASH Manifests vs. Segments

Segment TTLs and manifest TTLs require opposite treatment. Static media segments — once written, they never change — should carry 24-hour or longer TTLs. Serving a cached .ts or .m4s segment from edge memory costs nothing in origin egress; letting it expire and re-fetching costs exactly the same as a cache miss.

Live manifests (.m3u8 / .mpd) are the opposite case: TTLs of 2–6 seconds, aligned to your segment duration, prevent viewers from receiving stale playlists while keeping manifest requests hitting edge cache for as long as safely possible. Poster art, thumbnails, and init segments: treat these like static assets, cache aggressively, and strip versioning parameters from cache keys.

Origin Shield Placement and Its Effect on Egress Volume

Origin shield collapses cache misses from hundreds of edge PoPs into a single forwarding layer. Without it, 200 edge nodes each issuing independent cache misses on a new live segment generates 200 origin fetches for identical bytes — a fan-out pattern that both overloads origin and produces unnecessary egress.

Place the shield region as close to your origin infrastructure as possible, not at the geographic center of your audience. The goal is minimizing shield-to-origin round trips, not minimizing shield-to-edge hops. For a streaming architecture with origin in us-east-1, the shield should be us-east-1 — even if 60% of viewers are in APAC.

---

Encoding and Transcoding Costs: The Hidden Line Item in Your Video CDN Stack

What Multi-Rendition Encoding Actually Costs at Scale

Paid transcoding services bill per output minute, and the rate scales with resolution — a 4K rendition costs multiples of what the 480p rendition costs for the same source. A four-rendition ladder means you pay four times per minute of source, once for each output track.

Pull your own rate card and run the arithmetic: output minutes per event × number of renditions × per-minute rate for each resolution tier. The number that matters is annualized against your ingest schedule, because encoding cost compounds with event frequency, not with audience size. That property is what makes it easy to miss — encoding never shows up on the CDN line item, which is exactly why it's routinely excluded from TCO models that compare providers on egress alone.

Why Free Encoding Changes the TCO Equation

When encoding is bundled at no additional charge, the comparison stops being CDN egress vs. CDN egress and becomes CDN egress vs. (CDN egress + transcoding cost). For teams running weekly live events or high-volume VOD pipelines, bundled multi-rendition encoding — covering 4K through 480p — eliminates a line item that compounds with ingest frequency.

Convert it to a common unit to make the comparison honest: divide your annual encoding spend by your CDN's per-GB rate, and you get the volume of egress that bundled encoding effectively buys you for free. At $0.0049/GB, every $1,000 of eliminated encoding cost is worth roughly 200 TB of delivery. Factor that in and a bundled encoding tier changes the ranking at any meaningful ingest volume.

ABR Ladder Design: Balancing Quality and Bandwidth

A poorly designed ABR ladder inflates both encoding cost and storage without improving viewer quality. Four renditions where two mid-bitrate tiers are redundant — say, 1500 kbps and 1800 kbps both targeting 720p — consume encoding resources and storage for variants that ABR logic rarely selects. Each unnecessary rendition also adds a playlist entry, marginally increasing manifest size across millions of segment requests.

Design the ladder around actual network condition breakpoints in your viewer population. For a global audience with meaningful mobile share, the gap between 480p and 1080p is the most consequential quality jump — not the difference between two 720p variants. Fewer, better-spaced renditions reduce encoding cost, storage cost, and the storage egress generated when ABR clients request initialization segments for variants they abandon.

---

How to Migrate Your Video Streaming CDN: A Step-by-Step Cutover Playbook

Pre-Migration Audit: What to Document Before You Touch DNS

The most commonly underestimated phase. Configuring a new CDN means meticulously replicating every role, policy, and behavior of the primary via its API, followed by rigorous testing. (Source: IO River, 2025) That replication is impossible if you don't have a complete inventory of what you're replicating.

Document the following before touching a single DNS record:

1. Cache key configuration — which query parameters are normalized or stripped, which headers are included in cache keys
2. Signed URL / token auth patterns — HMAC algorithms, token TTLs, key rotation schedules
3. Origin shield topology — which shield region(s) are active, failover behavior
4. Header forwarding rules — which request/response headers pass through vs. get stripped
5. Custom edge logic — any Lambda@Edge, Cloudflare Workers, or equivalent compute running at the edge
6. Cache invalidation patterns — how cache purges are triggered (API, path-based, tag-based) and by which systems
7. Origin failover configuration — secondary origin URLs, failover conditions, retry logic
8. TLS certificate chain — whether certificates are managed by the CDN or externally provisioned

Parallel-Run Phase: Shadow Traffic and Cache Priming

Before any production traffic moves, prime the new CDN's cache by configuring it to pull from your existing CDN's cached responses — or from origin directly — while routing zero viewer traffic to it. Two approaches work:

1. Origin-pull warming: Point the new CDN's origin at your existing CDN hostname. It will pull and cache all requested segments through the existing CDN, building cache fill without viewer exposure. Disable after cache is warm.
2. DNS weighted routing (0%): Configure the new CDN distribution but assign 0% traffic weight at the DNS layer (Route 53 weighted records, or equivalent). Run synthetic load tests against the new CDN directly using realistic HLS playlist request patterns — not synthetic ping tests. Verify cache hit ratio is building correctly via the new CDN's analytics before incrementing any live weight.

DNS Cutover Strategy: Weighted Routing and Canary Percentages

Start at 5% weighted routing to the new CDN. Monitor for a minimum of 24 hours before incrementing. The metrics that must pass at each canary stage:

Do not increment stages during live events. If a live stream is scheduled within 72 hours, hold at the current stage until the event completes.

Post-Cutover Validation: The Metrics That Confirm Success

Within 48 hours of reaching 100% traffic weight, VOD cache hit ratio should exceed 85%. Live stream startup time should be within 10% of the pre-migration baseline. Segment TTFB from the new CDN should be consistently below 150ms at P95 across all regions you've migrated.

Pull origin request volume from your origin server logs independently — not from CDN-reported cache hit ratio alone. If origin request volume hasn't dropped proportionally to what the new CDN's cache hit ratio implies, there's a cache key normalization problem or an origin shield misconfiguration to resolve.

Rollback Protocol: How to Revert in Under 15 Minutes

Keep the previous CDN's distribution fully configured and live for a minimum of 72 hours post-cutover. Do not delete origin configurations, signed key pairs, or edge rules until you've confirmed the new CDN is fully stable.

Rollback procedure:

1. Set DNS weighted routing back to 100% on the previous CDN's hostname — TTL propagation is the bottleneck; set DNS TTL to 60 seconds at least 24 hours before cutover begins
2. Confirm traffic is routing back via CDN access logs within 3–5 minutes of DNS change
3. Validate origin request volume returns to pre-migration baseline
4. Do not attempt to diagnose the failure until traffic is fully recovered

---

Handling Traffic Spikes: How CDN Billing Models Break Under Live Event Load

Why Tiered Billing Punishes You for Success

A platform running a steady 50 TB/month baseline hits 160 TB in the month a playoff game lands. The extra 110 TB arrives in a single afternoon. On a committed-volume contract that incremental traffic is not billed down the tier ladder — it is billed at the overage rate, which is the one rate in the agreement that never goes down.

The bill arrives weeks later. There's no mechanism to cap it in real time. You find out the cost of success retroactively.

Commit-and-Overage Contracts: The Clause That Costs $40K

A team on a 50 TB/month commit at $0.07/GB (blended) pays $3,500/month at baseline. The playoff spike adds 110 TB in overage at $0.12/GB (a common overage rate): 112,640 GB × $0.12 = $13,517 in overage on one event. Do that twice in a playoff season and you've added $27,000 to a contract that was supposed to save you money.

The commit structure exists to give the CDN provider guaranteed revenue. The overage clause exists to monetize the upside of your audience growth. Neither clause benefits the buyer.

Flat-Rate Egress as a Spike Insurance Model

ZeroBuffer charges $0.0049/GB in every region, with volume tiers that only move downward — so a 10x traffic spike costs at most 10x the baseline, never more. There is no overage rate, because there is no commit to exceed. For the playoff scenario above, the same 112,640 GB of extra traffic costs about $552 rather than $13,517.

---

Choosing the Best CDN for Video: What the Benchmarks Actually Show

The Metrics That Matter for Video CDN Performance

Generic CDN speed tests — ping latency to an HTML endpoint — tell you nothing useful about video segment delivery under concurrency. The metrics that matter for a CDN for video workload are specific:

• Video startup time: Target <2 seconds at P95. Measures end-to-end latency from play() call to first frame.
• Rebuffer ratio: Target <0.3%. Measures percentage of playback time spent stalling. This is the metric viewers feel most acutely.
• TTFB for first segment request: Target <150ms. Drives startup time directly.
• Cache hit ratio: Target >85% for VOD, variable for live depending on segment duration and TTL configuration.

Global PoP Coverage vs. Latency: What 25ms Actually Means for Startup Time

A 25ms average round-trip to the nearest PoP means the first segment request completes in roughly 25–75ms depending on segment size and connection overhead — well within the <150ms TTFB target. But average latency is the wrong metric; P95 latency from your actual viewer geographies is what determines whether 5% of your audience gets a bad startup experience.

A 99.99% uptime SLA means less than 53 minutes of downtime per year. For live event delivery, read the SLA language carefully: "HTTP delivery" availability and "streaming delivery" availability are sometimes defined differently in the contract, with different remedies. A CDN that meets its HTTP SLA during a live event failure may still leave you without recourse.

How to Run Your Own Benchmark Before Committing

Use a realistic test setup: a real HLS manifest with 6-second segments, tested simultaneously from probe nodes in each region you serve, under concurrent load matching your expected peak viewer count. Single-client latency tests do not expose cache behavior under concurrency or the effect of origin shield on miss rates.

Run tests for at least 30 minutes — long enough for the CDN's cache to warm across its PoP fleet. Monitor cache hit ratio from the CDN's own analytics dashboard in parallel. Any provider that doesn't expose real-time cache hit ratio per PoP during evaluation is one that doesn't want you to see that number.

---

FAQ

What's the best CDN for high quality video streaming?

The best CDN for video streaming combines sub-25ms edge latency, cache hit ratios above 85% on VOD segments, region-independent egress pricing that eliminates bill shock at live-event scale, and global PoP coverage without regional surcharges for APAC and LATAM. On a pure TCO basis, providers that price the same in every region are the strongest option for teams serving global audiences or running live events, where tiered pricing resets monthly and overage clauses expose you to unpredictable costs.

---

What is CDN for video?

A CDN for video is an edge caching layer that stores HLS and DASH segments at PoPs geographically close to viewers, reducing segment fetch latency and eliminating the origin-fan-out problem that occurs when thousands of concurrent clients request the same bytes simultaneously. The architecture decouples origin capacity from audience scale — origin serves the CDN, not the viewers.

---

Can videos be served by CDN?

Yes, and serving video directly from origin at scale is an antipattern that collapses under concurrency. Every production video streaming CDN architecture routes viewer segment requests through CDN edge nodes. Serving a 50,000-viewer live stream direct from origin would require origin infrastructure capable of sustaining tens of gigabits per second of simultaneous outbound throughput — infrastructure that costs more to operate than any reasonable CDN egress bill.

---

Is there a free CDN?

Cloudflare's base tier provides free HTTP delivery but does not include video streaming features — Cloudflare Stream is a separately priced add-on billed per minute of video stored and delivered. Most "free" CDN tiers exclude media delivery, cap bandwidth below production-useful thresholds, or monetize elsewhere. The relevant question for streaming infrastructure isn't free vs. paid but cost-per-GB at your actual traffic volume. At scale, the difference between $0.085/GB and $0.0049/GB matters far more than whether a free tier exists.

---

Conclusion

Staying on a tiered, regionally-surcharged incumbent CDN is a defensible choice under specific conditions: your traffic is substantially contained within North America and Europe, monthly volume stays below 50 TB, and you have a committed support relationship that justifies the contract overhead. If all three of those are true, the status quo is probably not destroying your margins.

If you're scaling globally, running live events that spike unpredictably, or watching APAC and LATAM line items compound every month, the math in this article is not abstract. At 100 TB/month, the gap between CloudFront's blended $0.0705/GB and $0.0049/GB is about $6,700/month — $80,000/year — for identical bytes delivered. The engineering work to migrate is a one-time cost measured in days. The billing savings are permanent and begin on the first invoice cycle after cutover. Run the TCO formulas from the comparison section against your actual regional traffic split before your next billing cycle closes.

If the numbers in this article look familiar on your own invoices, ZeroBuffer's pricing page publishes every tier and a bandwidth comparison tool, so you can check the arithmetic in this article against your own volume. No sales call required to see the math.