Some thoughts from the inside. My company (Inbot) is currently in the pre ICO phase: we're selling tokens but not on the blockchain yet. We're very hesitant to move to the next phase precisely because of the tendency of many ICO companies to collapse almost straight away. Investors dumping your coin on the day of the ICO is kind of deadly for its valuation and if your ICO is attracting that kind of investors, something is indeed wrong. Quite many ICOs are set up such that this is actually the goal rather than something they want to avoid (pump and dump). We're kind of going against the trend here by planning to survive this initial phase (hard).
The reason many companies are doing ICOs regardless is because it actually makes sense if you look at it as a financial instrument that gets you funded quickly to the extend where you get similar capitalization comparable to a big series A or even B round.
Like with high risk investments, ROI is pretty bad; except when it is not. Most ICOs will fail but some will not. This space continues to be so hot on the premise that there are a few unicorn ICOs out there that are actually going to make it. Ironically, this is leading to a lot of misguided investments by, well, idiots. This in turn is leading to a lot of successful attempts to separate these idiots from their cash. Blame the idiots, not the ICOs. Hence Dogecoin, PonziCoin V3 (this is a thing !!!), and similarly poorly veiled scams. I applaud these people for being able to make so much money with so little effort.
There are many issues with dapps. IMHO it doesn't make any sense to write dapps right now unless your roadmap spans several years because the technology is very immature, the infrastructure is not there, and the fixes are years out from becoming proven to the point where you can rely on them to function as advertised. This requires more than proof of concept. Ethereum and bitcoin are out there with many users that are trying to exploit it in any way they can. That makes it battle tested. Alternative technology stacks exist but lack this type of user base and scrutiny.
Most of these alternative stacks still have serious flaws conceptually and practically. Quite a few of them seem to be moving the problem rather than solving it. What's needed is orders of magnitude improvements to throughput and scale for dapps to be usable by the masses. 50k transactions sounds nice until you realize that is globally and needs to service the entire ecosystem of dapp applications. That's not a lot compared to a modern db running on off the shelf hardware. Once users actually show up, that kind of traffic can be caused by a spike in usage on a single dapp. If you have many of those, that will be happening all the time.
Right now the only things dapps make sense for are things that are relatively high value so you can justify the cost and have enough per transaction revenue and very low in transaction volume so that the whole thing doesn't collapse when you actually get some traction. That pretty much narrows it down to limited amount of trading of coins on top of ethereum, which is indeed the dominant type of dapp in use today.
Some successful ICOs are actually about fixing this problem. They are long term investments in tech companies that are likely to fail but very lucrative if they don't. E.g. Telegram is a high risk investment but they are now well funded and have enough runway to execute whatever it is they are doing for the next few years. And they have a decent tech team. Similarly IOTA, QTUM, HashGraph, and others are well funded companies well capable of executing roadmaps that will take quite long to deliver results. Are those bad investments? Not if they succeed. Is that any different from e.g. the likes of Google bank rolling Magic Leap? Or indeed any kind of VC tech investment?