This just isn't true though. I worked at a relatively minor video streaming company and we overloaded and took down AWS CloudFront for an entire region. They refused to work with us or increase capacity because the datacenter (one of the ones in APac) was already full. This was on top of already spreading the load across 3 regions. We only had a few million viewers.

We ended up switching to Fastly for CDN. There's something hidden here though that becomes a problem at Netflix size. We were willing to pay the cloud provider tax, and we didn't dig down into kernel level or storage optimizations because off the shelf was good enough. At Netflixes scale, that adds up to millions of extra server hours you have to pay for if you don't do the 5% optimizations outlined in the article.