They don't want to be a Tier 1. They want to be as close to their users as possible, and the two are pretty antithetical goals. And while they could certainly afford to buy out one of the large networks, it would cost far more than what they are currently paying for connectivity, and they'd end up with a mostly unrelated business unit with thin margins and a lot of competition. While the bread and butter of these Tier 1s is Internet transit, the gravy is in the transport, advanced services, government contracts and so on. And these guys are big enough that they can squeeze the price wayyyy down to the point where the network is hardly making anything off them. They'd also become subject to a bunch of common carrier rules in various countries that they likely want no part of.
Being on a 'Tier 1' frequently results in quite sub-optimal paths, as traffic from the 9/10ths of the Internet not transiting through that particular Tier 1 will route via a major POP which may be 1000s of KM from the user. The Tier 1s only peer with each other in a handful of cities on each continent. They also do not peer with very many other networks, compared to a 'Tier 2', or what is available by connecting to a large peering exchange, which increases as-path length.
So while they have extensive dark fibre, some of which they have built, they would prefer to connect to users with direct peering - or even better, cache boxes within the end user's network - wherever possible, and owning a Tier 1 doesn't really help with that. In the case of a CDN, getting close to the user is exactly what they're selling. So they invest their money where it matters - installing cache nodes around the Internet and connecting to IXPs. CloudFlare for example has 165 POPs now. Google is at 167 peering exchanges. At any of them they will peer with anyone, so that's likely 2000+ ISPs they're directly connected to. Akamai claims they have cache nodes installed in over 2,200 locations.
Bandwidth is cheap, and buying a Tier 1 wouldn't necessarily help their goals, even if it saved them money (which is doubtful).
merci06-25 18:42
4楼
想到了一个同样的问题“当年Google分布式系统那么牛逼,为什么不先搞云计算”
Lemon06-25 18:47
5楼
@merci #4 这话说的没问题吧, 就是因为他看不上云计算, 所以 aws 比 google cloud 强的不止一星半点