They do that some times, including carefully measured network cables so every customer has the exact same latency. https://www.wired.com/2012/08/ff_wallstreet_trading/

The exchanges aren't monolithic institutions, they have numerous integrations with feeds from other exchanges and other order systems. Taking advantage of that disparity is one of the ways HFT techniques can profit.

Also, not every fund can fit their computers into an exchange building, nor would they necessarily want to.

This was discussed in the book "Flashboys". The real challenge with high speed trading in the long distance communication between the stuff on Wall Street and the commodities action going on in Chicago. That long distance communication is what many of these towers are being used for.

Is colo done in adtech as well? I imagine with more and more sophisticated customer modeling engines, each millisecond counts, and the closer you are to the exchange the more of those 100 ms you can spend on your own IP and now data transmission.