Email's a great example. You could set up your own email server, but doing so in such a way that you don't harm your ability to get emails into people's inboxes is problematic in all sorts of ways: you've got a good chance of being classified as spam. So even though the system is decentralized, the fact that most users are centralized within a few big players means that those parties can unilaterally impose rules that cut off users who don't participate in the centralized ecosystem.
In short: if Coinbase becomes sufficiently large, they could have the de facto power to disadvantage people who choose to use bitcoin outside of major centralized services.
This isn't to say that a system can't tend more or less to centralization: sure, it can, and Bitcoin could have a strong tendency to make unilaterally enforced centralization hard. But even good technical solutions are vulnerable to social forces, and social forces can be very powerful.
Git is decentralized, that means I can quit Github any time I want and use another service to host my repositories.
Bitcoin is the same: I'm not locked into Coinbase, I can switch them for another service, personal wallet on a phone, a hardware wallet, or a combination of them. Move my coins to a company in another jurisdiction etc.
Just the very fact of how easy it is to switch, creates a pressure on your provider and makes them work harder.
US banking is not decentralized. I cannot simply take dollars out to some other service or wallet. Moving them takes a lot of time, there is no privacy, there is suspicious reports and interrogation everywhere, civil forfeiture laws and for another jurisdiction, I have to sell dollars for another currency which incurs fees, risks etc.
In short: Bitcoin is like Git, fiat money and banking is like CVS.
You could move your open source project away from GitHub, but if GitHub is sufficiently popular that most people don't understand how the patch system works in git, and hence don't know how to contribute without using the affordances that GitHub offers, then you will incur a cost by doing so.
The point I'm making is that if the majority of people using a decentralized system via a centralized wrapper, and if interaction with those people is a priority, then it is possible for the provider of that centralized wrapper to make it harder for you to interact with their customers without yourself using the centralized wrapper. The more they do so, the less value there is to you (the decentralized user) in the millions/billions of people using decentralized service X through centralized wrapper Y, because they're effectively unreachable unless you also use said centralized wrapper.
This can happen by continuing to use the decentralized infrastructure, but penalizing participants who aren't accessing the infrastructure through a popular system. Alternatively, it can happen by having the decentralized infrastructure co-opted, so even though the service to some extent interacts with it, most user interaction actually exists entirely within the centralized system.
I mean, it doesn't seem crazy for me to imagine a world in which (eg) as a Coinbase user there's some monetary advantage to receiving payments from Coinbase members vs the wider Bitcoin world. At which point maybe merchants want you to pay via "Coinbase Send" (or w/e this hypothetical service is called) rather than pure Bitcoin transactions. At which point, sure, Bitcoin is the underlying unit of value, but no-one is actually using Bitcoin transactions (Coinbase is just moving around their internal accounting DB records), and hence the fact that things are counted in Bitcoin is basically a historical artifact.
I have no idea what will happen. My argument is just that I don't think that it is impossible for decentralized systems to be functionally neutered when people mostly interact with them through centralized wrappers.
Luckily, with Bitcoin, it's trade what you can lose. Keep the rest by yourself. Yes, there were some losses, but it's a trading platform. Somewhat expected.
If you don't believe me research what happened with the Swiss Franc recently. It has closed a lot of FX trading platforms from sheer losses [1].
This isn't limited to bitcoin — that's a naive and ignorant point of view.
[1] http://www.bidnessetc.com/32853-swiss-franc-shock-highlight-...
Apparently Coinbase now offers a multi-signature feature, so people could do the same.
The interesting thing here is that it's build on an open protocol rather than on legislation and so this I believe will at least give the opportunity to create all sorts of interesting solutions that are currently impossible to do within the current banking system because technology moves faster than legislation.
So even if we are talking centralization or clustering as I like to call it, we are not talking exclusion.
I think there is a real risk that we'll see bitcoin processing completely swallowed up by these big players.