One thing is deciding that IBM has a brilliant future and that the stock is a steal at the current price, and trading with someone who feels the opposite, another entirely to have thousands of transactions a second. Even the former transaction seems a bit zero-sum in that one of the people involved has the wrong idea and is going to either lose money or lose potential money.
That said, maybe I'm missing something - I don't know that much about HFT and stocks/finance in general, so I don't claim to have everything figured out.
If there are ways to make profit by trading at sub-100ms resolution, but a 1ms-resolution exchange is not any better at facilitating the outside-finance economy than a 100ms-resolution one would be, then it seems like HFT is solving problems of the exchange's own creation. It could even genuinely be solving those problems, but if they're problems that only arise in the context of extremely-high-granularity exchanges, and there is no practical benefit to such high time resolution, then why not just axe the problems?