Best European Alternatives to Fastly

Why people leave Fastly

Correct the assumption that brings most people here first. Fastly has a reputation as the developer-friendly, approachable CDN, and its prices are not approachable at all: $0.12 per gigabyte in Europe for the first tier, falling to $0.08 for the next ten terabytes. The free allowance is 100GB and a million requests, and the cheapest platform package is $1,500 a month.

That premium buys something specific. VCL and Compute let you run real logic at the edge, purges take effect almost immediately, and the observability is the best in the category.

If your team writes edge code and ships it several times a day, you are paying for a development platform and the delivery is incidental. If your team set up a cache four years ago and has not touched it since, you are paying twelve times Bunny.net's rate for a cache.

The second thing worth weighing is structural rather than commercial. On 8 June 2021 a valid configuration change by one customer triggered a bug shipped on 12 May, and 85 per cent of the network began returning errors.

Fastly detected it within a minute and had 95 per cent of the network healthy inside 49 minutes, which is a good response. The architecture that makes a change land everywhere in seconds is also the architecture that makes a mistake land everywhere in seconds.

• Twelve times the rate, for a feature set you may not use At $0.12 per gigabyte in Europe, a site shifting two terabytes a month pays around $240 before requests are counted. The same traffic costs roughly $20 at Bunny.net and $80 at KeyCDN. None of that is unreasonable if the money is buying edge compute and instant global purges. It is very hard to justify if the configuration has not changed since it was written.
• One control plane, one blast radius Fastly's selling point is that configuration and cache invalidation propagate globally in moments. The 8 June 2021 incident is what that property looks like from the other side: a single customer's legitimate change met a latent bug and most of the network failed at once. Spreading delivery across two providers, or picking one whose changes roll rather than land, is a different risk posture rather than a safer product.
• The legal mechanism has been rebuilt twice already Fastly, Inc. operates from San Francisco and certifies under the EU-U.S. Data Privacy Framework, which is the current lawful route for transatlantic transfers. It is also the third such route: Safe Harbour was struck down in 2015 and Privacy Shield in 2020. Relying on it is legal today and is a standing item on somebody's risk register, which a European supplier simply does not generate.
• You are buying a platform when you may need a service Fastly hands you the controls and expects you to use them. Rules are yours to write, WAF policies are yours to tune, and there is no equivalent of a security operations centre included in the price. For an engineering team that is an advantage. For an organisation whose website is important and whose engineers are busy, a managed European provider that writes and watches the rules is a better match for how the work actually gets done.

What you have to replace, not just match

Separate what Fastly is doing from what Fastly could do, because the bill is the same either way.

There is caching and delivery, which everything here does. There is edge logic — VCL, or Compute in Rust, JavaScript or Go — which is a programming environment rather than a setting. There is the purge behaviour that lets a newsroom change a front page instantly. And there is the observability: real-time log streaming into whatever you already run.

Gcore is the only provider here that answers the second properly. KeyCDN answers the third at a fraction of the price. Bunny.net answers the first cheaply enough to change the arithmetic entirely. Nobody matches the fourth, and if real-time log streaming is wired into your incident process, treat that as the thing to replace first rather than last.

The alternatives compared

How each alternative compares to Fastly

What actually breaks when you switch

Edge code is the migration, everything else is logistics. Count the VCL snippets and Compute services first: header rewrites and redirects re-express quickly in another provider's rules engine, while anything doing authentication, request signing or origin selection is a rewrite against a different runtime with different primitives. That estimate, not the traffic volume, determines whether this is a week or a quarter.

Log streaming is the second thing to rebuild, and it is the one that surprises teams. Fastly delivers logs in real time into S3, BigQuery, Splunk or wherever you put them, and incident response quietly comes to depend on that. Other providers offer analytics and periodic exports. Decide before the cutover whether you are replacing the stream or changing how you investigate.

Third, mirror the cache configuration rather than reimplementing it from memory. Surrogate keys, stale-while-revalidate behaviour, custom cache keys and origin shielding all have counterparts elsewhere with different defaults, and the differences appear as a cache hit ratio that quietly drops and an origin that starts working harder. Run both in parallel and compare the hit ratio before you move the last of the traffic.

Is Fastly really expensive, or is that just the list price?

The published rate is $0.12 per gigabyte for the first tier in Europe and North America, dropping to $0.08 for the next ten terabytes, with requests billed at about a cent per ten thousand. Enterprise customers negotiate below that, sometimes substantially, but negotiation requires volume and a contract.

The comparison that matters is against what the alternatives publish without negotiation. Bunny.net is $0.01 per gigabyte with a $1 minimum; KeyCDN is €0.04 with no commitment. For any organisation below the volume where a Fastly account manager will engage, the list price is the price and the gap is an order of magnitude.

The platform packages make the positioning explicit: the cheapest starts at $1,500 a month. That is a sensible number for a media company with an engineering team on the edge platform, and an absurd one for a business website.

What do I actually give up by leaving Fastly?

The edge runtime, first. VCL and Compute let you make routing, authentication and personalisation decisions at the point of presence rather than at the origin, and code written against them does not port. Gcore runs JavaScript, TypeScript and WebAssembly at its edge, which covers the common patterns, but a substantial Compute service is a rewrite.

Second, the observability. Real-time log streaming into your own systems is Fastly's quiet advantage, and teams build incident response around it without noticing. Every provider here has analytics; none has that.

Third, purge behaviour under pressure. A newsroom correcting a live front page cares about this in a way a brochure site never will. KeyCDN's instant purge is the closest equivalent on this page and it is genuinely close.

Does moving to a European provider make outages less likely?

Not by itself, and it would be dishonest to claim otherwise. Every provider here pushes configuration across a network, and every one of them can ship a bad change. Fastly's June 2021 incident is unusually well documented, not unusually bad — the company published the cause within a day, which most operators do not.

What changes is the correlation. When a large share of the internet sits behind the same control plane, a single fault takes down your site alongside your payment provider, your status page and half the services you would use to diagnose it. A smaller European operator fails on its own schedule.

If availability is genuinely the requirement, the answer is two providers with DNS-level failover, not a different single provider. That costs more and it is the only version of this argument that actually works.

Our compliance team accepted the Data Privacy Framework. Is that enough?

For today, it is a lawful basis and Fastly certifies under it, along with the UK extension and the Swiss framework. Nobody should tell you that using a DPF-certified processor is unlawful, because it is not.

The caveat is history rather than speculation. This is the third mechanism for transatlantic transfers: Safe Harbour fell in 2015 and Privacy Shield in 2020, in both cases after they had been relied on for years by organisations that had done nothing wrong. A European processor removes the question rather than answering it, which is why the question keeps appearing on risk registers.

The practical test is what happens to you if the framework changes. If the answer is a weekend of DNS changes, carry on. If it is a re-architecture, that is a reason to do the work at a time you choose.

Which one to pick

If your team ships edge code, Gcore is the only European answer on this page, and the decision is whether the rewrite is worth an EU-established supplier. If nobody has touched the edge configuration in two years, that decision does not arise and Bunny.net will do the same work for roughly a twelfth of the money.

If you chose Fastly for purge speed rather than for the platform, KeyCDN gets you that specific behaviour at €0.04 a gigabyte with Swiss law attached, and the rest of the platform was never the reason you were paying.

If the real problem is that nobody has time to maintain a programmable security layer, Myra Security replaces the tooling with people, which is a different purchase and often the correct one for a regulated organisation.

And if the answer to "who writes our edge code" is a team that does it weekly and well, Fastly remains a strong product. The premium is defensible when the platform is used. It is only indefensible when it is not, and the bill looks identical either way.