CDN as a Service vs. Building Your Own

Why this matters: almost every company should buy. Knowing precisely why — and what would have to be true to change that — is a genuinely senior judgment, and Netflix is the canonical worked example of the exception.

Key takeaway

Most companies avoid building their own CDNs, using providers like Akamai, Cloudflare, or Fastly, or cloud platforms such as AWS. Building only pays off at extreme, sustained scale — and even then the public CDN usually stays as backup.

The risks of depending on a public CDN

Companies contract with providers to distribute content. That introduces three specific risks:

Risk	Detail
Outages	Service failures are outside the content provider's control — you cannot fix them, only wait
Coverage gaps	If a provider lacks proxies in a region, you must contract additional services or deploy private infrastructure
Blocking	Shared domains or IPs may be blocked in certain regions because of banned content from other customers

Specialized CDNs

Some companies build specialized CDNs to control costs and performance. A specialized CDN uses points of presence (PoPs) dedicated solely to the company's content. These may function as caching servers, reverse proxies, or application delivery controllers.

Initial setup costs are high, but long-term operational costs are often lower than commercial services.

Private CDNs scale to serve petabytes and often coexist with public CDNs — the public CDN serving as backup during capacity shortages or failures.

Netflix Open Connect

Netflix's Open Connect Appliance (OCA) is a prominent example of a specialized video delivery CDN.

OCA servers do not store user data. Their primary tasks are:

Reporting status — health, routes, and cached content — to the Open Connect control plane in AWS.
Serving cached content directly to users.

The Open Connect operations team monitors all OCAs, whether at IXPs or embedded in ISP networks.

Why Netflix built it

Six reasons:

Reason	Detail
Scalability	Commercial providers struggled to scale infrastructure fast enough to match Netflix's growth
Cost	Rising streaming volumes made third-party services prohibitively expensive
Security	Custom infrastructure offers better control over data security and leakage risks
Control	End-to-end control over the video player, network path, and servers for optimal delivery
Diagnostics	Custom HTTP modules and TCP algorithms enable faster detection and troubleshooting of network issues
Retention	Cost-effective long-term retention of popular content, which is expensive with public CDNs

Netflix achieves a hit ratio close to 95% using OCA.

When building is right

Condition	Why it matters
Your traffic is a large fraction of global internet volume	Provider pricing stops reflecting your marginal cost
Your catalogue is bounded and demand is concentrated	A high hit ratio is achievable, which is what makes the economics work
You need end-to-end control of the delivery path	Custom transport and player-level optimization aren't available from a provider
Providers can't scale or reach where you need	The coverage-gap risk has already materialized
None of the above	Buy. The setup cost and ongoing operations are not worth it