Network delay has two components that behave completely differently, and confusing them is why people buy bandwidth to solve latency problems and are disappointed.
Transmission delay: bandwidth fixes this one
Transmission delay, also called serialisation delay, is the time to clock a packet onto the link. It is simply the packet size in bits divided by the link rate.
A standard 1,500 byte Ethernet packet is 12,000 bits. On a 100 Mbps link that takes 120 microseconds. On a 1 Gbps link it takes 12. This is the part of the delay that a faster connection genuinely improves, and it improves in direct proportion.
Propagation delay: bandwidth does nothing
Propagation delay is the time for the signal to travel the distance. It is distance divided by speed, and the speed is fixed by physics and the medium.
Light in optical fibre travels at about two thirds of its vacuum speed, because glass has a refractive index. That works out at roughly 5 microseconds per kilometre, which is the single most useful number in this whole area.
One hundred kilometres of fibre is 0.498 milliseconds one way, near enough half a millisecond, and about one millisecond for the round trip. Ten thousand kilometres, roughly London to Tokyo, is 49.8 milliseconds one way and 99.6 for the round trip. No upgrade reduces that. It is the floor.
Which one dominates flips with distance
On a short link, transmission dominates and bandwidth is the lever. On a long link, propagation dominates and bandwidth is nearly irrelevant.
Take that 1,500 byte packet over 10 kilometres of fibre. At 10 Mbps, transmission is 1.2 milliseconds against 0.05 for propagation: transmission is twenty times larger. At 1 Gbps, transmission falls to 0.012 milliseconds while propagation is unchanged at 0.05, and the ordering has reversed.
Past that crossover, buying more bandwidth barely moves the latency of a single packet at all. The transmission delay calculator reports which term dominates for any given link, which is the question worth asking before an upgrade.
Why your ping is worse than the calculation
Propagation is the floor, not the answer. Real latency adds queueing in routers, processing at every hop, and the fact that fibre routes follow railways and roads rather than great circles. A measured round trip two or three times the propagation floor is entirely normal.
What the floor tells you is which part of that is negotiable. If your measured ping is 95 milliseconds and the propagation floor is 90, there is nothing left to optimise. If the floor is 5, the other 90 are queueing and equipment, and those can be worked on.
The practical consequence
Latency-sensitive systems are placed near their users, not given faster links. Content delivery networks, edge compute and regional data centres all exist because the only way to reduce propagation delay is to reduce the distance.
Geostationary satellite makes this vivid: the orbit is about 36,000 kilometres up, so the round trip through the satellite is around 480 milliseconds before anything else happens. Low earth orbit constellations sit far closer, which is the entire reason they can offer usable latency.
For the distance component, see the propagation delay calculator. For memory rather than network timings, see the RAM latency calculator.