When reading about SpaceX Starlink, high frequency stock traders are mentioned as a potentially lucrative customers, this is because satellite to satellite communications don't have glass between them to slow the signal like regular cabled connections.

When will (or is it already?) satellite communications send a message from London to New York with less latency than ground-based communications? What are the milestones that Starlink needs to achieve to have superior latency? How much lower latency can Starlink be, given current plans for orbits, etc?

  • 3
    $\begingroup$ If you think regular cabled connections with glass fibers are always slower than satellite to satellite communications, you have misunderstood the limitations of both technologies. The link between ground and satellite has its own limitations of bandwidth. A fair comparison needs a lot of care. $\endgroup$
    – Uwe
    Jun 12, 2023 at 9:10
  • 2
    $\begingroup$ No, high frequency traders simply deploy their algorithm and computers in the same building, next to the stock exchange's servers. $\endgroup$ Jun 12, 2023 at 9:37
  • 2
    $\begingroup$ @user3528438 Some of the transatlantic cables were built just to shave off a few milliseconds of latency compared to the already existing ones. $\endgroup$
    – asdfex
    Jun 12, 2023 at 10:01
  • 2
    $\begingroup$ @user3528438 they do do it, however a sever in New York still sometimes uses information from London. An example for the ridiculous effort they will go to shave of some milliseconds is the privately funded secretly built cable from Chicago to new York $\endgroup$
    – moshevi
    Jun 12, 2023 at 11:00
  • 2
    $\begingroup$ Land connections frequently don't follow the geographically shortest path, meandering through "most bandwidth free" routes instead, often hamstrung by inter-business politics. I remember when my packets from one provider in my city in Poland were routed through Sweden and Chicago back to Cracow to another provider, because the two had a beef with each other and refused to create a local bridge. $\endgroup$
    – SF.
    Jun 12, 2023 at 13:42

1 Answer 1


Not Yet

As of early June 2023, satellite to satellite comms are not enabled on Starlink, which means currently packets hit the satellite and then return to a ground station for further on transmission. The long distance latency improvement you seek can only work if you cut out the ground section entirely.

And Probably Never

NYC to London one-way time is about 28ms. You can double check this with some simple math: 5,585 km / (0.66*300,000 km/s) = 0.028s

Where the 5,585km is the NYC to London distance and fiber transmissions occur at roughly 66% the speed of light in vacuum.

Starlink orbits at about 550km, which means we need to add 1,100km to the distance for the up and down legs - but we use the full speed of light so that helps.

6,700 km / 300,000 km/s = 0.022s

This looks good for Starlink, but there's also hop latency, and the likelihood of more hops.

Router Latency

Each router will add around 5ms of latency. The traditional fiber scenario has one router hop - the one at the remote side. The Starlink version will have 5: one on each satellite in each direction, and one at the ground on the remote side. So the round trip is now 28x2 + 5 vs 22x2 + 5 + 5 + 5 + 5 + 5 or 61ms for fiber and 69ms for Starlink. Already a problem for Starlink.

Additional Hops

This is also a problem. It is unlikely that Starlink will transmit from directly over NYC to directly over London. Rather, it is likely that multiple hops will occur in space as the curvature of the Earth and the presence of atmosphere prevent a direct laser link across 6,000km from LEO.

Each hop will add 10ms per round trip, and additional transmission distance (we're not moving in a straight line any more - instead we're zig-zagging)

Router Improvements and Bottom Line

It's possible to drive the router delay down (obviously not all the way to zero, but lower). So maybe the router delay can be driven low enough that it is negligible compared to the extra travel distance, but I'm not going to bet on it. If multiple satellite hops are required, the extra distance and extra router delay will both be meaningful penalties.

Starlink will probably never beat fiber for transatlantic communications latency.

  • 1
    $\begingroup$ @JCRM - you're right, I grabbed the round trip time and then treated it as the one way time. Updated the answer to show that Starlink is unlikely to ever beat fiber because the extra 1,000km and extra router hops are just too large a penalty to overcome. $\endgroup$
    – codeMonkey
    Jun 21, 2023 at 15:36
  • 1
    $\begingroup$ 5 ms is an extremely pessimistic estimate for routing latency. They should be able to forward packets in just microseconds, even without cut-through switching. $\endgroup$ Jun 21, 2023 at 16:21
  • $\begingroup$ @ChristopherJamesHuff Any traceroute will tell you that "microseconds" are extremely optimistic, at least with the routing hardware running the Internet these days. Hundreds of microseconds per hop might be just about achievable if you really put in a lot of effort. $\endgroup$
    – TooTea
    Jun 21, 2023 at 16:26
  • $\begingroup$ @ChristopherJamesHuff - I'm seeing ping times for Starlink quoted online as about 25ms (best case) and about 40ms (average). Since the space leg of Starlink is only 7.2ms round trip at the speed of light, I'd say router hop time is not negligible. $\endgroup$
    – codeMonkey
    Jun 21, 2023 at 17:27
  • 1
    $\begingroup$ @jcaron - there's a lot we don't know about this system. I'm really curious how they "aim" the optical laser link, given that many of the satellites are in constant motion relative to each other. The simplest way would be to aim at the satellites ahead of and behind you in the same orbit - but there's no way to do a NYC to London directly with that setup. You'd just go up or down the line until you find a good ground station and then route the rest terrestrial. So yah - there's a lot of refinement I could do with this answer, but until we know how the routing chain works it's not worth it. $\endgroup$
    – codeMonkey
    Jun 23, 2023 at 21:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.