Buying CDN

Buyer's guide.

Evaluation criteria

CDN evaluation is the discipline of comparing PoP coverage, performance, and pricing model together. The cheapest CDN is rarely the fastest for your traffic, and the fastest is rarely the cheapest at scale.

• Edge presence. PoP count and geographic coverage per vendor. Match to customer geography rather than the marketing map; presence in regions where you have no users is irrelevant.
• Performance. TTFB and tail latency from key customer regions per region. Run benchmarks before committing; vendor-published numbers are best-case.
• Pricing model. Per-GB egress, per-request, or flat-rate per vendor. Model your real volume; the cheapest model varies by traffic shape.
• Proof-of-value. Thirty to ninety day POC per vendor against a representative slice of traffic. Catches performance gaps and integration issues before purchase rather than after.

Major CDN options

Four CDNs cover most enterprise picks. Each fits a different shape; the choice is rarely about features alone.

• Cloudflare. Strong global presence with security bundled (WAF, DDoS, bot management). Predictable pricing; the default pick for multi-cloud or security-first orgs.
• Akamai. Long-established enterprise feature set. Higher cost, richer customisation; the right pick when the customisation surface actually gets used.
• AWS CloudFront. Tight AWS integration with per-region pricing. Best for AWS-heavy stacks where the integration savings exceed the feature gaps.
• Fastly. Edge-compute-oriented and engineering-friendly. Strong for teams that want to run logic at the edge rather than treat the CDN as a passive cache.

Hidden costs to ask about

Hidden costs are where the contract bites. Origin egress, WAF and security add-ons, and cross-cloud egress can each turn a cheap headline price into an expensive bill.

• Origin egress. Origin-to-CDN traffic policy per vendor. Some include this; some bill it separately at non-trivial rates.
• WAF and security add-ons. Separate line items per vendor. Bundle pricing matters; some vendors include WAF, others charge per request inspected.
• Cross-cloud egress. Cross-cloud CDN egress cost per flow. Can be expensive when origin and CDN sit in different clouds.
• Support tier. Included versus paid support per vendor. Catches the post-sale surprise when a Sev-1 needs a senior engineer and the contract only covers email response.

Performance evaluation

Performance evaluation is its own discipline. RUM, multi-region tests, and cache hit rate together cover the failure modes that any single metric hides.

• Real user monitoring. RUM dataset per CDN against real customers. Synthetic tests miss real-world variability in last-mile networks and ISP peering.
• Multi-region tests. Customer-geography tests per CDN. A CDN strong in North America may be weak in Asia; the marketing map does not show the gaps.
• Cache hit rate. Cache-hit gauge per CDN. The fastest CDN is meaningless if cache hit rate is low; misses fall through to origin and pay full latency.
• Failover behaviour. Origin-failover characteristic per CDN. Latent reliability gaps surface during the test, not the outage.

Decision framework

The decision is shape-driven. AWS-heavy, multi-cloud, enterprise, and edge-compute each point to a different answer; matching shape to vendor avoids years of operational friction.

• AWS-heavy stack. CloudFront pick per org. Integration savings (IAM, ACM, S3 origins) outweigh the feature gaps for most teams.
• Multi-cloud or security-focused. Cloudflare pick per org. Bundled security features and multi-region presence by default.
• Enterprise with custom needs. Akamai pick per org. Cost is real; capability matches when the customisation budget gets used.
• Edge compute focus. Fastly pick per org. The platform is designed for it; running logic at the edge is the supported path rather than a workaround.