It's in the article, but 25-35ms total. For the highest orbit they plan, one way latency is under 5ms, so these goals seem easily within the realm of possibility.
With multiple base stations that's the minimum latency to get across the street.
However, the added latency is not necessarily that bad vs landlines. Picture a triangle with the satellite at the apex and the goal at the other side. On the other hand if their base station was further from the network than your house then things get significantly worse.
With so many satellites in the sky, I think those effects will be exactly as you said, a few (<10 ms) milliseconds in the worst case, depending on how aggressive the hopping from sat to sat is.
Being able to break the comcast/att/verizon effective monopolies across the USA would (hopefully) have great positive effects. At least competition would bring prices down for low-latency land-based connections.
Of course, you're swapping control of your internet connection from a "mostly evil" to a "not yet known if evil" corporation.
Companies aren't "evil" as such, but they can get complacent when not exposed to sufficient competition. Once SpaceX is a viable competitor, Comcast will probably fairly quickly "forget" that they are evil.
I seem to recall people reporting that Comcast is actually just fine in many areas with a second viable provider. With SpaceX, that would be all of them.
There are the good looking Bond villains and the bad looking ones.
Obama and Musk are good looking Bond villains. (The former had assassination sky robots, ICBMs, and a military. The latter has rockets, oceangoing spaceship landing pads, and huge enigmatic installations in the middle of the desert.) Putin and Trump are bad looking Bond villains.
They'll just turn around, provide better service at lower prices, drive spacex out of the market, and raise them again. They have a lot more money to play with (and their infrastructure is much cheaper to maintain).
"They" meaning... every ISP on the planet? At the same time? Starlink will have the ability to offer coverage globally, even in places where existing ISPs have no infrastructure.
Not to mention that that kind of predatory pricing is illegal in many countries.
That's just the initial system. And there will be, say, 30 satellites in view at any one time, giving an aggregate of 600 Gbps. At a typical 100:1 over-subscription ratio, they should be capable of serving almost a million customers per region (~500km diameter) at that speed. At more reasonable 10 or 100Mbps, it's on the order of tens of millions of customers in a region.
But that's just the initial constellation. SpaceX plans to put 12,000 total satellites up, with the VLEO ones having much higher throughput. Idea is to replace them every 4-6 years with faster throughput.