A mid-market streaming company recently shared that their Fastly invoice jumped 38% quarter-over-quarter without a proportional traffic increase. The culprit was not bandwidth. It was request-based charges, regional surcharges in Asia-Pacific, and a log-streaming add-on they had forgotten was metered separately. This kind of surprise is not rare, and it is entirely predictable once you understand how Fastly CDN pricing actually works. This article gives you the full cost model: per-GB and per-request rates as of Q2 2026, the surcharges that do not appear in the headline number, a workload-profile decision matrix comparing Fastly to Cloudflare, CloudFront, and BlazingCDN, and a framework for estimating your real monthly spend before you sign anything.
Fastly bills on a pure consumption model. There is no flat-rate plan. As of May 2026, the published pay-as-you-go rates break down into two primary meters: bandwidth (per GB delivered) and requests (per 10,000 HTTP/HTTPS requests). Both meters vary by geography.
| Region | Bandwidth (per GB) | Requests (per 10K) |
|---|---|---|
| North America | $0.12 | $0.0090 |
| Europe | $0.12 | $0.0090 |
| Asia-Pacific | $0.19 | $0.0090 |
| South America | $0.28 | $0.0090 |
| Australia / New Zealand | $0.19 | $0.0090 |
These are the published list rates from Fastly's pricing page as of Q2 2026. Enterprise contracts with committed spend typically negotiate 20-40% discounts, but the regional multipliers remain proportionally similar. South America and APAC traffic is where invoices inflate fastest for globally distributed workloads.
The bandwidth and request rates are only the starting point. Several cost centers sit outside the headline numbers and regularly catch teams off guard.
Fastly charges for log streaming endpoints. If you ship logs to Datadog, Splunk, BigQuery, or S3, you are billed per million log lines. For high-request-volume APIs, this can exceed the cost of the bandwidth itself. A service handling 500M requests per month generates a meaningful log-streaming bill that does not appear in any bandwidth estimator.
Fastly IO is billed separately per transformation. If your front end serves responsive images across multiple breakpoints, each variant counts. At scale, IO charges for an image-heavy e-commerce site can add 15-25% to the base delivery cost.
Origin shield is free to enable, but the traffic it generates counts toward your bandwidth meter. This is architecturally sound since it protects your origin, but teams that assume shield traffic is "internal" and unmetered learn otherwise on the first invoice.
Custom TLS certificates with dedicated IPs carry a monthly fee. Compute@Edge usage (the successor to VCL-heavy custom logic) is billed per invocation and duration. If you run non-trivial edge logic, this is a third billing axis beyond bandwidth and requests.
Most Fastly pricing estimators only account for bandwidth. Here is a more realistic model. Assume a SaaS platform serving 50 TB per month, 80% from North America and Europe, 20% from APAC, with 400M requests and real-time log streaming enabled.
| Cost Component | Estimated Monthly |
|---|---|
| Bandwidth (40 TB NA/EU at $0.12 + 10 TB APAC at $0.19) | $6,700 |
| Requests (400M at $0.009/10K) | $360 |
| Log streaming (~400M lines) | $200-$400 |
| Origin shield bandwidth (est. 5 TB) | $600 |
| Total | ~$7,860-$8,060 |
The same 50 TB on BlazingCDN's volume tier would cost $350 per month for the first 100 TB, plus nothing additional since 50 TB falls within the included allocation. Even doubling to 100 TB keeps the invoice at $350. At 500 TB, BlazingCDN's price is $1,500 per month with additional GBs at $0.003. For enterprises pushing petabyte-scale delivery, rates drop to $0.002 per GB at the 2 PB tier. This is the kind of cost structure, with stability and fault tolerance comparable to CloudFront, that makes BlazingCDN's volume pricing worth modeling against any Fastly or CloudFront quote. Sony is among the enterprises already running production traffic through BlazingCDN at these rates.
No single CDN wins every workload. This matrix maps four common delivery profiles to the provider that best fits each one as of 2026.
| Workload Profile | Best Fit | Rationale |
|---|---|---|
| High-traffic API with complex edge logic (VCL/Wasm) | Fastly | Sub-second config propagation and Compute@Edge make Fastly strongest where edge programmability is the primary requirement and budget is secondary. |
| Flat-rate web app delivery with integrated security | Cloudflare | Cloudflare's Pro/Business plans bundle delivery and security at fixed monthly rates. Predictable budgets, less flexibility on cache logic. |
| AWS-native architecture with S3/Lambda@Edge | CloudFront | Zero egress from S3 to CloudFront and native integration with the AWS ecosystem make CloudFront cheapest when your origin is already in AWS. |
| High-volume media/video or software delivery (50 TB+) | BlazingCDN | At $0.004/GB (25 TB tier) dropping to $0.002/GB (2 PB tier), no other provider matches the per-GB cost for bandwidth-heavy, latency-tolerant workloads at scale. |
Yes, and the mechanism is straightforward. Fastly has no built-in spend cap on pay-as-you-go accounts. If a product launch, viral event, or DDoS reflection drives a 10x traffic spike, your invoice scales linearly. In Q1 2026, Fastly introduced usage alerts via the dashboard, but these are notifications, not circuit breakers. By the time the alert fires and a human reads it, thousands of dollars in delivery charges may already be accrued. Teams that rely on Fastly for bursty workloads should implement their own rate limiting at the application layer or set up automated origin failover to a lower-cost provider for overflow traffic.
Fastly offers a free developer account with limited traffic. As of May 2026, the free tier includes up to $50 worth of traffic per month, roughly 400 GB of North American delivery. This is enough for staging environments and proof-of-concept work. It is not enough for any production workload. The free tier does not include Compute@Edge, Image Optimization, or dedicated TLS, so evaluating Fastly's full feature set requires a paid account.
As of Q2 2026, Fastly charges $0.12 per GB in North America and Europe, $0.19 in APAC and ANZ, and $0.28 in South America on pay-as-you-go. Request fees are $0.009 per 10,000 requests across all regions. Enterprise contracts with committed spend can negotiate lower rates.
The primary advantage is zero commitment: you scale to zero when traffic drops, and there is no minimum term. The primary disadvantage is unbounded cost exposure during spikes, compounded by separately metered services like log streaming, image optimization, and Compute@Edge invocations that inflate invoices beyond bandwidth projections.
Cloudflare's Pro ($20/month) and Business ($200/month) plans include unlimited bandwidth with no per-GB metering, making them far more predictable for bandwidth-heavy sites. Fastly becomes more competitive when you need instant config propagation, VCL-level cache control, or Compute@Edge. For pure delivery cost, Cloudflare wins at virtually every traffic level.
Fastly's official pricing page provides a basic bandwidth calculator but does not include log streaming, IO transformations, or Compute@Edge costs. To estimate accurately, you need to inventory every billable feature you plan to use and model each one separately. The framework in this article provides a starting template for a more realistic projection.
At 100 TB per month, BlazingCDN's tier price is $350 flat. Fastly's list rate for the same volume in North America would exceed $12,000 before request and add-on fees. CloudFront with committed-use pricing lands around $7,000-$9,000 depending on region mix. For pure bandwidth cost at high volume, BlazingCDN is the lowest-cost option with production-grade reliability.
Pull your last three months of Fastly invoices. Break out bandwidth, requests, log streaming, and any Compute@Edge or IO charges into separate line items. Calculate your effective per-GB rate by dividing total invoice by total GB delivered. If that number exceeds $0.15, you are paying a significant premium over what the market offers in 2026. Model the same traffic through two alternative providers using their published rates, then run a one-week parallel test on your staging environment to validate latency and cache-hit ratios before making any migration decision. The data will tell you more than any vendor pitch.