Full order book feed is nice but realistically, there very little benefit in paying for it unless you have the infrastructure to play low latency games. The feed itself is fairly affordable, especially if you concentrate on a single-listed product (for example, CFE full book feed is 3500 a month and touch-only is 1500). However, low latency setup that can take advantage of the alpha coming from the atomic order book updates is probably gonna run you a few hundred grand at least (especially if you include developers into that cost). Most large funds that are not involved in market making do not bother with the direct feeds and instead pay for something like RT or Bloomberg. That sounds about right if one wants to compete in the top tier, though from what I know about his business he is active across multiple venues which adds a lot of costs. You can get away with much less if you concentrate on very specific products/strategies. Anyway, I was actually wondering if he's talking about non-latency sensitive stuff like venue access.
That figure ($120K) is probably at least double today. HFT is a race to zero, and its impossible to keep up with the big boys (Citadel, Jump, Optiver, Susquehanna) who will always outspend and always be faster than you. Although low latency helps, my trading system, which is part automated (black box), part old school "point-and-click" doesn't need that kind of speed. I have my own server co-located in Aurora (CME data center), but not a direct feed from any exchange. And right now we're all on the same playing field trading over the internet.
Should have known you'd be aware of that stuff! My bad. But I think the info that feed provides could be useful for manual trading trading slower moving (maybe especially an extra thick market like ZN where there's a lot of stuff happening in the queue?) contracts on short term time horizons (I know garachen has manual traders and does some manual trading himself, I think they use the full order book data feed features for that too, but maybe not) so I don't think it necessarily has to be only for really low latency strategies. But maybe I'm wrong. I couldn't see it being very useful for a manual trader trading on short term time horizons (seconds) with a really fast moving contract which is what I'm working on. Too much info to process in a timely manner and people and algos often trading 1-2 lots at a time with less action in the queue because the books relatively thin. TT actually provides queue position for CME products (queue position is provided by the exchange, not just estimated like with some platforms) and I know a lot of people manually trading the ZN find it really useful but with faster markets it's not as useful. And you'd have to spend a lot of time with the data, I doubt there's anything obvious there (talking about the info provided from the direct data feed) just subtle context dependent clues that you have to train yourself to see and understand so I get that it's no free lunch. And yeah the 120,000 is for data and colocation etc on multiple exchanges and for multiple traders.