At the risk of being contrarian and wrong (just another form of dumb in my book) I do often wonder whether there isn't a distributed systems bet that is the anti cloud hedge in the next generation of computing. So while Amazon stores more and more of the data in the world and delivers most of the MIPS in a metered way (or by the request with Lambda) and client devices become less stateful and more reliant on the network, someone somewhere figures out some applications that actually benefit from being totally distributed.
The classic two cases have been those that solve for network latency (storing big files closer to where they are going to be consumed and user control (or government control in the case of China. The first feels like a moment in time as both wired and wireless broadband gets faster and the second is likely to persist for a while, at least while governments continue to dictate Internet policy. My favorite latest project here is Zeronet— though admittedly still far too geeky, it seems like the right was to deal with censorship.
Earlier today my friend Andy mentioned another anti-cloud bet when I brought this up: tracking the state of many users in fully simulated world. In a limitation that I did not know about, most high performance game servers seem to top out around ~150 players. If this is true, there may just be the first reason to create a giant highly performant distributed state machine that could track infinite virtual worlds without needing commensurate Amazon bills.
Worth thinking about, especially as we are currently barreling simulated worlds from many different vectors (VR, IoT, 3D design just to name a few...)