CDN Cost Economics: 95th-Percentile, Commit, Overage, Transit

Why this matters

If you run a video product, your CDN line is the largest variable cost item on the engineering side of the P&L. Bitmovin's Video Developer Report 2025/26 names cost control as the number-one challenge for 38% of video teams — more than any technical concern. A team that does not understand its CDN bill loses negotiating leverage, makes the wrong build-versus-buy calls on origin shielding and multi-CDN, and routinely ships pricing decisions to the procurement table without the numbers to defend them. The reader for this article is the engineer who has been asked "why did the CDN bill double last month?" and the product or finance lead who is signing the contract. By the end, both should be able to read a CDN invoice line by line, project next quarter's bill from a viewership forecast, and tell a sales rep whether the commit they were just offered is good or bad.

The four meters on a CDN bill

Before you can compare two CDN quotes, you have to know that they are quoting against four meters, not one. The headline rate — "$0.02 per gigabyte" — is only the first.

The first meter is data transfer out (also called egress, sometimes "data transfer to internet"). It is the count of bytes the CDN ships from the edge to viewers, measured in gigabytes per month, and it is the single largest line on most streaming bills. CloudFront's published pay-as-you-go rate for North America and Europe begins at $0.085 per GB after the first free terabyte each month, drops to $0.080/GB for the next 40 TB, and continues sliding down to $0.020/GB at petabyte scale. South America starts at $0.170/GB — exactly twice the US/EU rate.

The second meter is HTTP requests. Every manifest fetch, every segment download, every key request is a request that the CDN counts. For most websites this meter is rounding error, but a live HLS player makes one playlist request every few seconds plus one segment request per rendition for every segment, and that adds up. CloudFront prices US/EU requests at $0.0075 per ten thousand HTTP requests and $0.0100 per ten thousand HTTPS requests — HTTPS is roughly 33% more expensive than HTTP across every region. The first ten million HTTP(S) requests each month are free.

The third meter is peak bandwidth, expressed in Gbps. Pure CDN contracts rarely meter this directly, but enterprise deals — and almost every IP-transit or content-peering deal that sits behind a CDN — bill on the 95th percentile of a 5-minute sample. We will dissect this meter in the next section because it changes how you think about traffic shape entirely.

The fourth meter is origin egress — the bytes your origin sends to the CDN to fill its caches. If your origin is in AWS S3 and your CDN is CloudFront, AWS waives this. If your origin is in S3 and your CDN is Cloudflare or Fastly, you pay AWS's standard egress rate of $0.05–$0.09 per GB on every cache miss. This is the line that surprises teams switching providers, because their old bill hid it and their new bill does not.

A handful of CDNs charge additional small meters on top — HTTPS request invalidation past a free quota, edge compute invocations, image transformations, log delivery, regional uplift fees, real-time log streams. Treat these as the long tail. A clean, large streaming workload's bill rounds to the four meters above plus a flat "support / SLA" minimum.

Figure 1. The four meters on a streaming CDN bill — every quote you compare must line up against all four, not just the per-GB rate.

95th-percentile billing, from first principles

95th-percentile billing — also called burstable billing — is the meter that confuses new buyers the most, and the one that decides every IP-transit contract and every premium CDN private deal. The mechanics are simple; the intuition takes a worked example.

The provider samples your bandwidth in 5-minute windows for the entire month. A 30-day month has 8,640 such samples. At the end of the month, they sort the samples from highest to lowest, discard the top 5% — roughly 432 samples, or about 36 hours of traffic — and bill you for the highest remaining sample. That number is your "95th-percentile bandwidth", expressed in Mbps or Gbps, and it is multiplied by your contracted price per Mbps to produce the bill.

The intuition is that 95th-percentile billing lets you burst above your committed rate for short windows without paying for the peak. If you run a live event that doubles your traffic for one hour every Saturday, those eight or so 5-minute samples a month sit safely inside the discarded top 5%. You pay for the second-busiest hour of your month, not the busiest.

Now do the math out loud.

A streaming product averages 3 Gbps for most of the month. On three evenings, it spikes to 12 Gbps for 90 minutes each as a big match goes live. Each spike contributes 18 five-minute samples, for a total of 54 spike samples. 5% of 8,640 samples is 432. The 54 spike samples are well inside the top-5% discard. So:

• 95th-percentile bandwidth ≈ 3 Gbps (the steady-state rate).
• At $1.00 per Mbps per month — a representative IP transit price for mid-volume customers in 2026 — the bill is 3,000 × $1.00 = $3,000/month.
• If the spikes were instead 4 hours long each, they would contribute 144 samples, still well under 432, and the bill would still be about $3,000.
• If a single ten-hour spike happens, that is 120 samples — still under the discard — but if it recurs every other night, it is 1,800 samples per month, well over the discard, and the 95th-percentile bandwidth shifts toward the spike level.

The lesson: 95th percentile is generous to infrequent bursts and punishing to regular ones. A weekly sports event is free; a nightly news show that doubles your traffic at 8pm is not.

The same meter is hostile to a workload with uniformly high utilization. If you saturate 10 Gbps every hour of every day, your 95th-percentile is 10 Gbps and you would have paid less on a flat per-GB price.

The takeaway most teams miss: the choice between per-GB and 95th-percentile is a choice about your traffic shape, not your traffic volume. Per-GB billing favours steady, predictable workloads (subscription VOD, software updates). 95th-percentile favours sharply peaky workloads (live sports, breaking news, game launches). The wrong meter for your shape can change the bill by a factor of two or three at the same total bytes.

Figure 2. Two workloads, two meters — the per-GB meter rewards the steady shape, the 95th-percentile meter rewards the bursty one.

Commit, overage, transit: three contract structures

Once you understand the meters, you can read the three contract structures every CDN sales team will quote you against.

Pay-as-you-go is the structure on every CDN's published pricing page. You ship traffic, the meter counts it, you pay the rate-card price at the end of the month. CloudFront, Cloudflare, Fastly, Google Cloud CDN, Azure Front Door, BunnyCDN — every provider has a public pay-as-you-go tier. The rate sits at the top of the volume-tier ladder; it is the worst per-GB price the provider offers. It is the right structure only when you genuinely cannot predict your volume more than a quarter ahead.

Committed-use ("commit") is the structure every CDN sales rep will steer you to once you spend more than a few thousand dollars a month. You agree to pay for a minimum number of GB (or a minimum dollar amount) per month for a fixed term — typically 12 or 36 months — and in exchange the rate drops. The discount versus pay-as-you-go is usually 25–60% at mid volume and can exceed 80% at hyperscale. Direct CDN contracts with Fastly, Akamai, or Cloudflare typically beat AWS CloudFront standard pricing once your bill reaches $30,000–$50,000/month. The two unobvious rules of commits: the unused portion of the commit is forfeit at the end of each month, and the overage rate for traffic above the commit is set in the contract and is usually 10–50% higher than the committed rate.

Worked example. You commit to 500 TB/month at $0.012/GB on a 24-month deal, with an overage rate of $0.018/GB. Your minimum bill is 500,000 × $0.012 = $6,000/month, payable whether you used the bytes or not. If you ship 700 TB in a heavy month, you pay $6,000 + 200,000 × $0.018 = $9,600 — which works out to $0.0137/GB blended. If you ship only 300 TB in a light month, you still pay $6,000, which is $0.020/GB blended. The economics work when your expected monthly volume is at or slightly above the commit; they collapse if you over-commit and routinely under-use.

IP transit ("burstable" / 95th-percentile) is the structure used by ISPs, peering exchanges, and a handful of CDN-adjacent providers (BunnyCDN sells a 95th-percentile tier; many cloud providers will quote you 95th transit if you ask). You pick a committed information rate (CIR) in Mbps — the floor — and a burst rate that is the technical maximum your port can push. You are billed monthly on the 95th-percentile of your actual usage. If the percentile is below your CIR, you pay for the CIR. If it is above, you pay the per-Mbps overage rate on the difference. The math from the earlier section applies: a peaky workload at 3 Gbps 95th-percentile with a 1 Gbps commit at $0.50/Mbps pays $500 minimum and $1,000 in overage at $0.50/Mbps for the 2 Gbps over commit — $1,500 total.

The three structures coexist on most large streaming stacks. The pay-as-you-go meter is the rate-card sanity check. The committed-use contract is the primary CDN spend. IP transit is what your origin is connected to — the cost of the bytes that move from your origin out to the CDN's first-tier nodes, which is often invisible until you change CDNs or move regions.

A fourth, smaller structure shows up on some pricing pages: the flat plan with bundled allowances. Cloudflare's Pro, Business, and Enterprise plans, AWS's Security Savings Bundle, BunnyCDN's High Volume tier, all bundle some amount of bandwidth and requests into a monthly subscription fee. AWS introduced the Security Savings Bundle in late 2023 and offers up to 30% discount on combined CloudFront + WAF spend for a 1-year commitment. AWS expanded its top-tier Premium flat plan in May 2026 to support configurable usage allowances up to 600 TB and 6 billion requests per month per distribution. These plans are good for SMB workloads that fit inside their allowances; they collapse to commit-plus-overage math for everyone above.

Figure 3. The four contract structures, with the bill-producing formula written out beside each one.

The big-five pay-as-you-go rate cards in 2026

The headline rate is not the whole story, but it is the starting point. The table below is the May 2026 pay-as-you-go rate card for the five large general-purpose CDNs that streaming teams most often shortlist. All prices are per gigabyte, US/EU region unless noted, for data transfer to internet.

Provider	Free tier	First-paid tier	Next tier	Discount tier	Notes
Amazon CloudFront	1 TB / mo	$0.085 (next 9 TB)	$0.080 (40 TB)	$0.060 (100 TB) → $0.020 (5+ PB)	South America 2× US/EU; APAC 1.2–1.5× US/EU
Cloudflare	Generous on Free / Pro plans (no hard byte cap)	Enterprise contract only above plan limits	n/a	Custom rates at scale	Egress to internet not metered on Pro/Business; metered on Enterprise
Fastly	100 GB trial	$0.12 first 10 TB	$0.08 next 40 TB	Custom below ~$0.02 at PB scale	Per-region multiplier table on pricing page; request fees additional
Google Cloud CDN / Media CDN	10 GB / mo	$0.08 first 10 TB	$0.055 next 140 TB	$0.030 above 500 TB	Cache-fill from non-Google origin billed as egress
Azure Front Door	100 GB / mo	$0.083 first 10 TB	$0.075 next 40 TB	$0.040 next 100 TB	Standard vs Premium tier multiplies the sheet

Notes the table cannot fit. The "list price" on every CDN is the worst price the CDN sells. Cloudflare's Enterprise plan does not appear on a public price page at all — it is custom-quoted, and the per-GB-equivalent for streaming customers typically lands in the $0.005–$0.015 range. Akamai (not on the table because it has no published pay-as-you-go rate for media delivery) is generally 2–4× Cloudflare and 1.5–2× Fastly at comparable traffic, but it is also the only one of the five with a media-delivery product (Adaptive Media Delivery) that includes per-segment cache-tag invalidation and a dedicated low-latency live workflow in the base price.

Two specialist CDNs are worth knowing for the streaming category specifically: BunnyCDN's High Volume tier and CDN77's media-delivery plans are both built around streaming and routinely quote $0.001–$0.005/GB at petabyte scale, plus 95th-percentile options. BlazingCDN publishes a $0.002/GB price at 2 PB commitment that is the cheapest publicly listed CDN price as of Q2 2026.

The headline number to walk away with: published rates span more than 40× from CloudFront pay-as-you-go ($0.085/GB) to BlazingCDN at petabyte commit ($0.002/GB). Most production deals after negotiation land in the $0.005–$0.030/GB range. If a vendor quotes you outside that range without explaining why, you are probably looking at either the wrong tier or a price that hides a different meter (origin egress, requests, peak Gbps).

Worked model: a 100,000-MAU live-and-VOD streaming product

Pricing pages are abstract. Bills are not. To make the meters concrete, walk through the math on a realistic mid-market streaming product.

The product has 100,000 monthly active users. Each watches an average of 5 hours per month, half live and half VOD, at an average bitrate of 4 Mbps (a 1080p stream after H.265 encoding and ABR adaptation). The live windows are concentrated: 80% of the live hours happen during a 4-hour weekday-evening window and a 6-hour weekend window. The product runs on HLS with 4-second segments and a 6-rendition ladder. Cache hit ratio at the CDN edge averages 92%.

Bytes shipped. Total user-hours = 100,000 × 5 = 500,000 hours. Bandwidth-hours = 500,000 × 4 Mbps = 2,000,000 Mb-hours = 2,000,000 × 3600 / 8 = 900,000,000 MB = 900 TB. Call it 900 TB per month.

Bytes from origin. With 92% cache hit ratio, origin egress = 900 × (1 − 0.92) = 72 TB.

Request count. A live HLS player on a 4-second segment cadence fetches ~6 manifest requests per minute and ~6 segment requests per minute, for ~12 requests per minute per viewer. VOD is similar after the initial buffer. 500,000 hours × 60 minutes × 12 requests = 360 million requests/month. Round to 400 million to account for retries, key fetches, and CMCD reporting. Assume 100% HTTPS.

Peak bandwidth. Concurrent peak during the busiest evening hour — assume 25% of MAU active in that hour and 80% of them watching live = 100,000 × 0.25 × 0.80 = 20,000 concurrent. At 4 Mbps per viewer = 80,000 Mbps = 80 Gbps.

Now plug those numbers into the three contract structures.

Scenario A — CloudFront pay-as-you-go in US/EU.

• Data transfer: 1 TB free + 9 TB × $0.085 + 40 TB × $0.080 + 100 TB × $0.060 + 750 TB × $0.040 (approximating the 350 TB and 524 TB tiers blended) = $0 + $765 + $3,200 + $6,000 + $30,000 = $39,965 in data transfer.
• Requests: 400M × $0.0100 / 10,000 = $400 in HTTPS request fees.
• Origin egress: assume origin is in S3 same region → CloudFront → free for the CloudFront leg. Origin egress within S3 is included.
• Total: ~$40,400/month. Effective per-GB ≈ $0.044.

Scenario B — Negotiated CloudFront commit at 800 TB/month.

• Negotiated rate, typical for this tier: $0.018/GB committed, $0.025/GB overage.
• Data transfer base: 800,000 × $0.018 = $14,400. Overage on 100 TB: 100,000 × $0.025 = $2,500. Subtotal $16,900.
• Requests: usually negotiated to $0.005/10k = $200.
• Total: ~$17,100/month. Effective per-GB ≈ $0.019. A 58% saving versus pay-as-you-go.

Scenario C — Specialist CDN on 95th-percentile transit.

• Quote: $0.50/Mbps for committed 50 Gbps; overage at the same $0.50/Mbps per Mbps above commit.
• 95th-percentile bandwidth in this workload: with live concentrated in 4×5 + 2×6 = 32 hours/week of peak (96 samples × 4 weeks ≈ 384 samples — comfortably under the 432-sample discard), the 95th-percentile sits near the non-peak traffic level: roughly 30 Gbps.
• Bill: 30,000 Mbps × $0.50 = $15,000. No request fee on transit. Origin still needs to be fed; assume internal.
• Total: ~$15,000/month. Effective per-GB ≈ $0.017.

The point is not that scenario C wins — it usually wins for this exact traffic shape, but it also requires the operational maturity to run a CDN-like distribution layer on raw transit, which most teams should not do. The point is the spread: the same 900 TB workload pays $40,400, $17,100, or $15,000 depending purely on contract structure. The headline per-GB rate of CloudFront pay-as-you-go is 2.4× the negotiated CloudFront rate and 2.7× the transit rate.

The next question every CFO asks is whether multi-CDN saves more money. We will answer that in a section of its own.

Figure 4. The same workload, four contract structures, four very different bills — the per-GB headline tells you almost nothing about the real cost.

Multi-CDN economics: when the spreadsheet says yes

A multi-CDN architecture routes user requests across two or more CDN vendors. We covered the architecture and failure modes in 6.4 — multi-CDN architecture and the standards-based way to implement it in 6.5 — content steering. Here we focus on the money.

The reasons to go multi-CDN are not all financial. Resilience (no single point of failure across the delivery layer), regional performance (one CDN is the fastest in Brazil, another in Vietnam), and contract leverage (each vendor knows you can shift traffic in a quarter) are all valid on their own. But the financial case rests on three specific mechanics:

The incremental commit mechanic. Most CDNs price the next thousand terabytes lower than the first thousand, so splitting traffic 50/50 between two vendors and committing to half on each can produce a lower blended rate than committing all of it to one vendor — but only if both vendors are willing to quote you at the same per-GB rate they would at full volume, which experienced sales teams refuse. Net: the incremental-commit argument usually loses, and the multi-CDN spend is the same or slightly higher than single-CDN at the same volume.

The overage avoidance mechanic. If you commit to 500 TB on each of two CDNs and ship 900 TB total, you pay base rate on both 500 TB tranches and no overage on either, because each CDN sees only 450 TB. A single 1,000 TB commit on one CDN at the same per-GB rate would have left you with 100 TB of overage. This is the most defensible financial reason to split traffic — but it requires steerable, real-time-switchable load balancing across the two CDNs, which adds operational cost and an additional vendor (the content-steering server itself).

The regional optimization mechanic. CloudFront's South America rate is 2× its US/EU rate. A regional specialist like Limelight (Edgio), G-Core, or CDNetworks may quote you a meaningfully lower per-GB rate for South America or for emerging APAC markets specifically. Sending South America traffic to the specialist and US/EU traffic to CloudFront can cut the regional component of the bill in half. The break-even depends on the absolute size of the regional traffic — if South America is 3% of your bytes, the saving is small; if it is 30%, the saving pays for the multi-CDN operational overhead by itself.

A useful rule of thumb from real deals we have seen: multi-CDN reliably saves 10–25% on the all-in CDN bill once monthly spend exceeds about $50,000, before any reliability or performance benefit. Below that line, the operational and content-steering costs usually eat the saving.

Origin egress: the line nobody quoted you

The bill line that breaks the most spreadsheets is origin egress. The pattern: a team that has always run on CloudFront with an S3 origin sees the AWS S3 → CloudFront leg as free (it is — AWS waives that transfer). They switch to Cloudflare or Fastly to save 40% on the per-GB CDN rate. The first month's bill arrives, the CDN line has indeed dropped, and a brand-new line — AWS Data Transfer Out — is suddenly several thousand dollars.

The mechanics: every cache miss on the new CDN goes back to the origin to pull the segment. Each pull is an HTTP GET that AWS bills at the standard S3 egress rate (currently $0.05–$0.09/GB depending on region and volume). At a 92% cache hit ratio on a 900 TB workload, that is 72 TB of origin egress. At $0.085/GB, that is $6,120. At $0.05/GB after S3 volume discounts, still $3,600.

Three corollaries.

First, the cache hit ratio on your new CDN is now a directly billable line item, not just a performance metric. A drop from 92% to 85% on the same workload adds $0.05 × (900 − 765) TB = $0.05 × 135 = ~$6,750 in additional origin egress — roughly the same magnitude as the entire savings you got by switching CDNs. We covered why hit ratios drift in 6.2 — origin shielding and tiered caching and 6.3 — cache keys for streaming.

Second, the origin shield product is a cost lever, not just a reliability one. A well-configured CloudFront origin shield can lift hit ratio by 5–15 percentage points by funnelling all edge misses through a single regional cache that fills from origin once. The shield adds a per-GB markup on the shielded path (CloudFront's is ~$0.0075/GB) but the math almost always wins: a 10-point hit-ratio improvement on a 900 TB workload at $0.05/GB origin egress saves 90 TB × $0.05 = $4,500, against a 90 TB × $0.0075 = $675 shield uplift. Net $3,825 saved per month.

Third, co-locating origin and CDN inside the same cloud is a financial decision as much as an operational one. CloudFront + S3, Google Cloud CDN + Cloud Storage, Azure Front Door + Blob Storage all waive the origin-egress meter on the same-cloud path. Hosting your origin in AWS and your CDN at Fastly costs you AWS egress on every cache miss. Hosting your origin at a neutral provider (Backblaze B2, Wasabi) buys you a different egress rate — Backblaze's B2 → CDN egress is free for the major CDN partners via the Bandwidth Alliance.

How to read a CDN quote

You will receive at least three CDN quotes during a real procurement cycle. Every quote will be structured to make the headline per-GB rate look as low as possible while keeping the bill flexible upward. The seven-line checklist below is the one you walk through before you sign.

1. What is the committed per-GB rate at my projected volume, on the exact region mix I will use, after all discounts? Not the next tier up; not the global average.
2. What is the overage rate above commit, and is it a flat multiplier or graduated? Buyers who negotiate overage rates at contract time typically secure rates within 10–20% of the committed rate; buyers who do not routinely face 30–50% overage premiums.
3. What is the request-fee schedule, HTTPS specifically? At a 4-second segment cadence on a 6-rendition ladder, request fees are not a rounding error.
4. What does origin-egress to your CDN cost, and from which origins is it waived? If you are not on the same cloud, this is a non-zero line.
5. What is the price of origin shield / tiered caching, and is it included in the per-GB rate? A shield product behind a paywall may still be the right answer financially.
6. What is the SLA, the credit formula, and the exclusion list? Most CDN SLAs credit 5–15% of the affected period and exclude force majeure, third-party network failures, and routing changes you initiated. A shaped credit formula is a sign of a serious provider; "best effort" is not.
7. Term, true-up cadence, exit clause. A 36-month commit at a 50% discount may look like a steal until your business shifts and the unused commit becomes a write-off. A 12-month with quarterly true-up at a 30% discount is usually the better deal.

Buyers who walk into procurement with these seven questions get better contracts than buyers who walk in with a "$/GB" target. The reason is structural — every CDN rep is a percentage of margin, so they steer the conversation to the line item where they have the most room. Anchoring on a single line lets them concede there and inflate the four lines you did not ask about.

Two pitfalls that quietly inflate the bill

Pitfall — letting cache hit ratio drift unmonitored. The CDN's cache hit ratio is the single most leveraged number on the bill. A drop from 95% to 85% on a 1 PB workload at $0.05/GB origin egress is a $5,000/month increase. The usual causes are cache-key drift (URL parameters added by an analytics team), TTL changes (engineering shortening manifest TTLs for "freshness"), and origin response headers that include no-cache. Pin hit ratio as a billable KPI in the monthly review, not an SRE-side metric.

Pitfall — committing to a 36-month deal at flat volume on a growing product. Streaming products do not grow linearly. A 36-month commit set to today's volume looks fine for the first six months and then either (a) becomes a massive overage line as you outgrow it, or (b) becomes a stranded commit as you migrate part of your traffic to a multi-CDN partner. A staggered structure — commit to the first 12 months at conservative volume, with pre-negotiated price points at higher volume bands for years 2 and 3 — costs more on day one and saves more on day 366.

A third one that is worth a sentence: never accept a quote that does not separate the request-fee meter from the data-transfer meter. A handful of providers will quote you a single "all-in $/GB" and absorb requests, and that quote looks great until you realise you are subsidising other customers' request-heavy workloads on a per-segment-heavy live-event-skewed traffic shape.

Figure 5. Matching workload shape to contract structure — the wrong corner of the chart costs 50–200% more than the right one for the same bytes.

Where Fora Soft fits in

Across our streaming, OTT, WebRTC, e-learning, telemedicine, surveillance, and AR/VR engagements, the CDN bill is almost always the single largest infrastructure line by the time the platform hits its first million viewer-hours. We design the cost model into the platform from the early architecture review: cache-friendly URL conventions and segment versioning that hold the hit ratio above 90% as the catalogue grows, origin-shield configuration that protects the egress line, multi-CDN steering when the geography or scale earns it, and a contract worksheet that converts the next year's viewership forecast into a defensible commit. The same scaffolding has saved several of our OTT and live-event clients high-six-figure annual CDN spend by replacing a default pay-as-you-go invoice with a properly negotiated commit-plus-overage structure on the same provider, before any architectural change at all.

What to read next

• Multi-CDN: the architecture, the cost story, the failure modes
• Content steering: the standard way to do multi-CDN
• Origin shielding and tiered caching

Call to action

Talk to a streaming engineer about modelling your CDN bill before the next renewal · See our case studies in OTT, live events, and telemedicine · Download the CDN Cost Cheatsheet — a one-page reference with the four meters, the three contract structures, the seven-line quote checklist, and the rule-of-thumb breakeven numbers used in this article.

References

1. Amazon Web Services. Amazon CloudFront Pricing. https://aws.amazon.com/cloudfront/pricing/ (accessed 2026-05-24). Data-transfer-out tiers by region; HTTP vs HTTPS request rates; 1 TB / 10M request always-free tier.
2. Amazon Web Services. Interpret your AWS bill and usage reports for CloudFront. https://docs.aws.amazon.com/AmazonCloudFront/latest/DeveloperGuide/billing-and-usage-interpreting.html (accessed 2026-05-24).
3. Amazon Web Services. CloudFront Security Savings Bundle. AWS news release, late 2023; expansion of Premium flat plan announced May 2026 — configurable allowances up to 600 TB / 6B requests per distribution.
4. Cloudflare. Cloudflare Pricing and Plans (Free / Pro / Business / Enterprise). https://www.cloudflare.com/plans/ (accessed 2026-05-24). Enterprise per-GB equivalent is custom-quoted; not a published number.
5. Fastly. Fastly Pricing. https://www.fastly.com/pricing/ (accessed 2026-05-24). Per-region multipliers and request-fee schedule.
6. Fastly Engineering Blog. How to Reduce Egress Traffic Costs with a CDN. https://www.fastly.com/blog/how-to-reduce-egress-traffic-costs-with-a-cdn (accessed 2026-05-24).
7. Google Cloud. Cloud CDN and Media CDN pricing. https://cloud.google.com/cdn/pricing (accessed 2026-05-24). Cache-fill billed as standard egress when origin is non-Google.
8. Microsoft Azure. Azure Front Door pricing. https://azure.microsoft.com/en-us/pricing/details/frontdoor/ (accessed 2026-05-24).
9. Bitmovin. Video Developer Report 2025/26 (9th annual). https://bitmovin.com/video-developer-report/ — "Cost control is the number-one challenge for 38% of video teams."
10. CTA (Consumer Technology Association). CTA-5004: Web Application Video Ecosystem — Common Media Client Data (CMCD). https://cdn.cta.tech/cta/media/media/resources/standards/pdfs/cta-5004-final.pdf (accessed 2026-05-24). The CMCD specification used by hit-ratio-aware steering decisions.
11. SVTA (Streaming Video Technology Alliance). Common Media Client Data — wiki entry. https://wiki.svta.org/common-media-client-data/ (accessed 2026-05-24).
12. Wikipedia (RFC-adjacent reference). Burstable billing. https://en.wikipedia.org/wiki/Burstable_billing — formal derivation of the 95th-percentile rule and 5-minute sample window standard.
13. Stackscale Engineering Blog. 95th percentile metering and billing of bandwidth. https://www.stackscale.com/blog/95-percentile-metering-billing-bandwidth/ (accessed 2026-05-24). Detailed worked examples of CIR vs burst sizing.
14. Noction Engineering Blog. 95th percentile and other bandwidth metering methods. https://www.noction.com/blog/95th-percentile-explained (accessed 2026-05-24). Practitioner overview of transit-pricing structures.
15. Akamai Engineering Blog. Cache hit ratio: the key metric for happier users and lower expenses. https://www.akamai.com/blog/edge/the-key-metric-for-happier-users (accessed 2026-05-24).
16. CacheFly. What engineers should know about reducing cloud egress fees in streaming. https://www.cachefly.com/news/what-engineers-should-know-about-reducing-cloud-egress-fees-in-streaming/ (accessed 2026-05-24).
17. BlazingCDN. CDN price benchmark: comparing 12 providers by GB served. https://blog.blazingcdn.com/en-us/cdn-price-benchmark-comparing-12-providers-by-gb-served (accessed 2026-05-24). Published-rate comparison up to PB-scale commit pricing.
18. CDN77. Media-delivery plans and 95th-percentile pricing options. https://www.cdn77.com/pricing (accessed 2026-05-24). Specialist streaming CDN with published transit-style tier.