I would not expect anything like AlphaZero do much in terms of alpha. There is a variety of reasons why deep learning has very little application to trading (aside from maybe image analysis). However, considering that Google holds all sorts of private data for all sorts of companies, that's a much bigger edge should they decide to exploit it. Of course, the moment they do that they will lose both that business and the information that comes with it. A retail guy actually has an advantage when it comes to systematic trading since his liquidity requirements are much lower and his mandate is literally "do anything that can make money". This said, if I were a retail guy, I'd not bother with anything latency sensitive, there are plenty of other stuff to do that's less technologically involved.
That’s true. You can grossly outperform the benchmark by simply rebalancing a portfolio once a month. This said, if you can develop the proper analysis and execution platform to feed that tick data into, you can do much better. A pipe dream for most retail, since they’re constantly looking for the path of least resistance. Been there, done that. No free lunch
Tick data usually runs anywhere from $50/m to $1k+/m. It really depends on the vendor and which feeds are included. As well as the amount of symbols you're looking to stream (real-time). Be sure to account for the latency of the real-time data when backtesting, if it's a latency sensitive strategy.