Not sure which bit you're replying to, but with this constellation latency is expected to be < 100 ms, more likely on the order of 35 ms.

I've heard reports of the test satellites achieving 25 ms, but haven't found a source yet.

The reason of course is that these satellites are in low earth orbit, so only a few 100 km away, rather than geostationary orbit like most internet satellites that are at 35 000 km away

Starlink, not having the speed of light slowed down by fibre, should have lower latency over long distances.

26ms lower from new York to London (total 50ms).

Past instances of this sort of speedup have shown that this is worth billions to high frequency traders...

Reddit post that includes includes links to the academic paper that latency number is coming from, and lots of other useful information: https://www.reddit.com/r/SpaceXLounge/comments/9sy4q7/starli...

This shouldn't be down voted. While photonic-bandgap fiber can approach speed of light in a vacuum (to ~99.7%) it is not commercially deployed in any volume AFAIK, and standard fiber propagates signal at only about 70% of vacuum. Starlink will of course involve a minimum fixed distance penalty of a few hundred kilometers on each leg and some amount of inherent constellation bounce penalty each sat. Even so over distances of a couple thousand kilometers or more having the signal go near maximum (even up/down, air is not vacuum but with an index of refraction of only 1.0003 is damn close) vs 70% actually looks to be quite significant. I see no reason while standard fiber couldn't be beaten even under optimal point-to-point conditions (like NY to London, around 5500 km) if they were able to get it up fast enough in the near term.

And that's point-to-point from an HFT viewpoint, whereas of course in most situations fiber will not be even close to taking a direct route between locations. Even if photonic-bandgap fiber got deployed for transoceanic or continental backbone links Starlink could still of competitive practical latencies if the satellites are orbiting low enough and impose a low enough overhead.