10,000 * 20Gbps = 200Tbps, and that all assumes that the traffic is going in and out directly from customer to internet with none of that capacity used for anything else like groundstation-to-groundstation hops or anything like that.
You'll find a *lot* more than 200Tbps of usable bandwidth in even a moderately sized national operator. At a small ISP (~90k users) we have about 25Tbps of capacity in our equipment.
A single mid-range carrier router can easily handle a dozen terabits; something like a Juniper MX2020 has 80Tbps capacity. A Nokia 7950 XRS-20e has 96Tbps. These are single racks worth of kit that deliver half the entire constellation's capacity.
Headroom is particularly important for usable high speed services as traffic becomes burstier. Statistical multiplexing gets you so far.
Now bear in mind that's with 10,000 perfectly working satellites and no degradation from 20Gbps based on spectrum availability and I'd be surprised if SL carries more than a million or two people - maybe 10 at the highest level.
Don't get me wrong, service for the hardest-to-reach million or two places on Earth with good latency and throughput would be awesome, though I think they have a lot of work to do on their impact on ground based scientific astronomy if they really want to launch and maintain 10,000 sats. But a mass-market service it ain't.
AIUI per-satellite bandwidth is power constrained.
The beauty of a five-year lifespan for the satellites is that when, for example, multilayer PV panels (i.e., more efficient/higher output) get cheap enough, it'll be simple to roll them out.
Likewise with more power-efficent routing chips, and everything else.
It's not 10,000 satellites. Starlink has FCC support (with the ITU) for over 40,000 satellites.[1]
With a steady-state demand of about 10k/year, Starlink will be able to get fairly good prices from semiconductor fabs for ASICs. Is the market for Juniper MX2020s the same size? (I honestly have no idea.)
No doubt Starlink will be researching furiously to optimise transmit power, modulation schemes, antenna design, et cetera as well.
A final point is that Starlink's markets are additive, not mutually exclusive.
They can do low-latency links over the North Atlantic and North Pacific for HF traders/arbitrageurs, and when the satellites are over other parts of the globe, they can serve village WISPs, or individual consumers (and the US military if you insist). The one market Starlink can't serve well is exactly that which is best served by fibre: urban areas in the developed world.
Looking at the business case, the risk is all upside for Starlink.
On 15 October 2019, the United States Federal Communications Commission (FCC) submitted filings to the International Telecommunication Union (ITU) on SpaceX's behalf to arrange spectrum for 30,000 additional Starlink satellites to supplement the 12,000 Starlink satellites already approved by the FCC.
Note that the cost of an MX2020 likely exceeds Starlink's cost of building and launching one of their satellites.
Now there is likely to be an relatively few MX2020s in a mid-size carrier (conceding that Starlink will be launching many more satellites) but Starlink has essentially 0 cost infrastructure for last mile deployment other than the satellites themselves.
But lets say Starlink gets to your upper limit of customers (which I assume they will, eventually, assuming they aren't already up against a Shannon limit on their bandwidth,) that is a billion dollars a month at $100 vs ~1.2 billion in launch revenues total in 2020, not even counting high revenue customers like the Department of Defense.
This is going to be a money printing machine for SpaceX once their main cost is just replacing deorbiting satellites and bandwidth on the ground.
Sure those big routers are costly but they're maybe the cost of a dozen satellites and serve half the total constellation size. Buried infrastructure has a 20-30 year guaranteed lifespan if built with fibre. Routers tend to last 10 years minimum. The opex cost of constantly replacing satellites is going to be pretty substantial.
Also Starlink doesn't have last mile costs as such (they just push that to the consumer) but they do have ground station requirements which will grow linearly with demand. It'll be interesting to see how they manage that as they grow. And of course, they're going to need those big routers too...
Will any one ground link station actually require the big routers though? The traffic will be distributed through each connection point to possibly push the throughput per ground station to a lower level of cost hardware.
It wouldn't surprise me if they put ground stations in every country. Doing a satellite to satellite hop across 10 countries needlessly wastes the limited bandwidth of your satellites. It makes more sense to just use the satellites for last mile delivery and hard to reach places.
You'll find a *lot* more than 200Tbps of usable bandwidth in even a moderately sized national operator. At a small ISP (~90k users) we have about 25Tbps of capacity in our equipment.
A single mid-range carrier router can easily handle a dozen terabits; something like a Juniper MX2020 has 80Tbps capacity. A Nokia 7950 XRS-20e has 96Tbps. These are single racks worth of kit that deliver half the entire constellation's capacity.
Headroom is particularly important for usable high speed services as traffic becomes burstier. Statistical multiplexing gets you so far.
Now bear in mind that's with 10,000 perfectly working satellites and no degradation from 20Gbps based on spectrum availability and I'd be surprised if SL carries more than a million or two people - maybe 10 at the highest level.
Don't get me wrong, service for the hardest-to-reach million or two places on Earth with good latency and throughput would be awesome, though I think they have a lot of work to do on their impact on ground based scientific astronomy if they really want to launch and maintain 10,000 sats. But a mass-market service it ain't.