For nieve signal distances, that can sometimes be true. That's not how starlink works however. It bounces the signal between satellites, each adding latency. Overall, fibre wins in almost every situation.

The bigger problem is saturation. Most things you can apply to radio waves can be applied to light in a fibre. The difference is you can have multiple fibres on the same run. This massively increases bandwidth, and so prevents congestion.

Just checked the numbers. Starlink is up at 550km. That means a minimum round trip of 1100km. In order to beat a fibre run, you are looking at over 2000km distance. Even halving that to (optimistically) account for angles, that's still a LONG run to an initial data center.

Satellites need to orbit at some distance above the planet though, so the round trip will always be fairly long even for ones with a pretty close orbit.

This makes no sense on the face of it. Let's say the satellites are 100 km (or miles) above the earth. If I was to connect to a server 10 km (or miles) away, my complete route over fiber is 10 km. My complete route over satellite is just over 200 km (assuming it's between those two points). Now, let's say the server is 500 km (we'll assume the earth is flat over this expanse, even though that's about 5° around the earth). So our fiber link has to go 500 km, more or less. Our satellite link has to go about 540 km, best case scenario. If we raise those satellites, it only gets worse (it's probably closer to 860, best case scenario, for satellites at 350 km).

I just did a quick check, and the curvature of the earth over that 500 km scenario is about 20 km (it won't be 20 miles for 500 miles).

Now, you might start to argue that were talking about straight lines, and that's true for satellites but not for fiber. And that might be true. But we've already shown that the hop to space and back is already increasing that distance by 60% or more. But those two or so straight lines are just til you get to the Starlink hub, so you aren't going to reduce this much more than the numbers above. And yes, fiber will have some extra distance due to following the grid rather than straight lines. But, again, that only matters to the ISP hub and then you're back to the same distances.

The other argument you listed is the speed of light in space/atmosphere vs. fiber, and it's a valid point. Not there are some interesting things done with guiding light to the center of the fiber, which is another way of saying there are multiple refractive indexes, but let's go with a refractive index of 1.5. That means the speed of light in glass is 2/3×c, or that light in space can go about 50% farther. And that's about the added distance for using LEO satellites.

tldr: All the benefits of transmitting through air or space are basically negated by the added distance, where the best-case scenario is only slightly better than the worst-case scenario for fiber.