When you serve traffic at this volume, it's no longer the requests per second that matter. You can easily have static content on lots of metals, but the new pr

When you serve traffic at this volume, it's no longer the requests per second that matter. You can easily have static content on lots of metals, but the new problem is saturation of peer links on egress, or unintentionally triggering DDoS mitigations along the path that the traffic takes (or on your own or the CDN services). That the content could be placed on enough metals to serve it is the easy part... but a nicely designed solution for serving the requests isn't as impressive as keeping the network solid and operating smoothly. This is also why in the tweet thread that performance, optimisation of resources, etc is called out explicitly... fail to do this and you kill your network. Fastly did great here, but the gov.uk people and GDS also did great in making the job for Fastly a lot easier.