I'm still learning about this industry and particularly satellite construction. I went to chat GPT to get a breakdown of costs. Amusingly it told me that Starlinks cost an estimated $250k and use as their satellite bus the Dragon Spacecraft which costs $300M. Anyway, it appears, and quite possibly equally wrong, that power, solar panels and electronics are around $40M for a GEOSAT, the Bus is $50M and the "Payload" is $50-$150.
If a starling is $250k, that's a HUGE savings, for what you call a "crap" satellite, but the failure rate for Starlinks over 5 years seems to be (anecdotally) around %5? Of course I am comparing Apples and Oranges.
To be clear I'm not calling the SL sats crap, I'm saying they fall WAY short of the bar from the perspective of a traditionally spec’ed satellite. And also to be clear, that's not a slight--SX has simply changed the way the bar gets constructed. Traditionally the number of nodes in a space system is so low (including N=1) such that, to turn the phrase, failure of any particular node really isn't an acceptable option. SX has simply created so many nodes that the reliability/lifetime concern fully shifts from thinking about the actual health of that satellite to solely thinking about the statistical health of that constellation--which to date is basically unheard of. (Even the existing big constellations like Iridium generally need all satellites operational, and failures are a big deal) That logic makes a lot of sense to all of us (I assume) but it does come at a cost that's unfortunately hard for non-space folks to understand and at a cost that’s pretty much untenable until you get to mega-constellation territory. Many many hundreds of sats, at least.
So...yes, a traditional GEO sat costs hundreds of millions of dollars, but you also can't compare the cost of a traditional sat directly with the cost of a starlink sat. You need to go up to the service level to do the comparison. Maybe the best case study right now is Viasat, who are building 1+ TBPS sats for their Viasat-3 constellation.
Bear with me as I'm going to do way more numbersing than anyone should be forced to endure here...for anyone who has the good sense to skip the following, BLUF is that the financials for a Starlink-like constellation are anything but an obvious slam dunk.
Ok, so conservative estimates puts the cost of a Viasat-3 at $650M, they're built to pump over 1 TBPS reliably for 15 years, and of course they need to be launched. That ends up rounding to a cost of ~$50k/GBPS/year. The constellation is actually 3 satellites with near global coverage, but let's just play out a hypothetical where one of those sats just puts down 1TBPS over the continental US.
Starlink sats can each pump 20GBPS, are built to last less than 5 years, and it's reasonable enough for this comparison to say that to put 1TBPS down on the lower 48 the constellation needs ~1000 sats (~50 usefully operating sats over the US at any moment). For the sake of keeping this 0th order, that roughly means 3000 sats over 15 years. Taking your $250k/sat cost for Starlink (which is maybe a little high based on what I've heard), and without factoring in launch costs, that maths out to the ~same $50k/gbps/yr as the GEO. Now of course there's the whole rest of the world where the constellation can also provide service (whereas the GEO can't in this hypothetical). To factor that out--as folks who at all follow satellite services know--the US will for the foreseeable future represent at least 50% of the global revenue of something like Starlink. So for the sake of simplicity we'll halve the $50k rate to $25k/gbps/yr for Starlink.
But...now we need to factor in launch costs. For SX to hit the $50k number they're looking at the equivalent Falcon launch cost of $6-7M (and we generally believe the cost is closer to $10M). If this were a non-SX company someone is looking at ~60 launches (@50 sats/launch) which would rack up to the better part of $3B just in launch costs (round number WAG of $50M/per) which skyrockets that annualized per bit number to $250k/year.
Note that the above doesn't even begin to contemplate other program costs like NRE or the ground segment, both of which are significantly higher for the megaconstellation over the GEO. Many tens of percent of the total program cost.
Now, of course there's a number of upsides to the LEO constellation, both in performance and robustness over the GEO sat (technology advancements, how the capacity can actually be used/maximized, etc.), but again the point here is that its not simply a slam dunk to swing the pendulum way into a megaconstellation over more traditional solutions. You gotta dig deeper for the SX angle to make sense.
Telesat is another relevant case study here. They were looking at ~300 sats @ ~$10M a piece, with something like ~18-20 launches. Basically, they wanted a comparable-ish constellation to Starlink, but because they didn't have internal at-cost access to launches, they traded constellation cost for launch cost: they decided to make the individual sats bigger, more capable, and more robust [than Starlinks] in order to reduce the total number of launches.
That logic racks up to around ~$4B for the space segment, or ~$250M/year (over 15 years, which I think was the goal) for the constellation. I'm pretty sure they could put more than 1 TBPS down on the lower 48--maybe even twice that--but if we go with the above 1tbps capacity that's puts the bean counters back at the ~$250k/gig/year number. And clearly investors don't like that number...
But I'm wondering is where does the money go to get "lifetime, reliability, and performance" required for traditional satellites?
Excellent question.
It's one of those deals where there's no one thing that makes a difference but rather the sum of a bunch of complimentary elements. What it really comes down to though is providing acceptable service quality and acceptable service uptime to a customer. Traditionally in a space application this is achieved with operational confidence of each of a low number of satellites; that confidence only comes with money. Megaconstellations achieve quality/uptime through the strength in numbers approach. (Non-megaconstellations, unfortunately, end up leaning toward the traditional approach.)
Certainly not comprehensive, the money for traditionally spec’ed satellites goes to:
—Radiation resiliency: Any electrical component on orbit is going to be the recipient of a reasonably predicable amount of radiation, and those dosages will [statistically] result in reasonably predictable failures, ranging from simple, correctable errors (like a bit flip) to hard failure. Generally speaking low orbits see significantly less radiation than higher orbits, so a satellite in LEO is going to need a much less robust set of electronics than something in MEO/GEO, for instance. (there's a lot of nuance in the actual dosages depending on what kind of orbit, the actual altitudes, etc.) Regardless, a component that's more resilient is going to cost more money. Often Significantly more money.
—Operational Lifetime: Very related to radiation but worth separating out because its a big factor, radiation is a cumulative thing, and any particular part has a reasonably accurate [statistical] life based [mostly] on dosage. A starlink chip that’s designed to last 4-5 years or whatever only sees half the dose of an otherwise equivalent chip on another program that wants an 8-10 year mission, and will cost maybe 10-25% the cost of the 8-10 year chip.
—Redundancy: Bit Captain Obvious here, but for a number of different reasons you want to have a spare thing in case the primary thing stops working. This could be an intermittent issue, like a computer resetting due to a bit flip (and so you want a backup computer to take over), or this could be permanent, like a component hard-failing and having to be bypassed. This applies at both the part level (like parts on a circuit board) as well as the top level satellite (where you might have 2:1 computers, 3:2 star trackers, 4:3 wheels, etc.). On top of these extra parts/units you need additional satellite resources (switches, harnessing, automated recovery logic, cross-strapping logic, etc.) to make it all work, which all costs more money, more time to develop, and more time to validate/test…and then of course all that adds complexity so you need to layer on even more to make sure that complexity is sufficiently controlled within your risk profile. All that costs money…and time and mass (which both cost money).
—Stable performance over mission life: Kind of related to the first two, you need your gizmos to do The Thing throughout the entire mission life. We’ve talked about radiation above; thermal is the other major player here. A Satellite goes through a lot of thermal cycles—that radio shack resistor, for instance, probably is going to be all over the place over temperature whereas a more expensive resistor will be acceptably stable. This phenomenon degrades over time, so it’s very related to mission life. Maybe that radio shack part works for a 6 month cubesat…but absolutely won’t for that 15 year GEO.
—Gizmo Pedigree/Traceability: This one adds a ton of money. You need to know That Part is the right part and that it’s going to work, and all that takes time and oversight to make sure it’s a valid part. Circuit board components are a great example here. Often a space rated part is actually a physically different part (materials, build processes, suppliers, etc.) than the radio shack equivalent in order to achieve the above points, so there’s obvious price increase there. But sometimes the space rated part is literally just a better controlled radio shack part (or at least, the automotive grade part)—things like lot traceability, lot screening, positive asset control, etc…all those cost money. The logic is that there’s variability in any mass produce part, and you don’t want to risk your umpteen million dollar satellite on getting unlucky with that one 1/10 cent resistor that didn’t get built right. Better to buy a 20 cent version where you’re confident it’s going to work. The fun irony in this is that the automotive part is getting built in massive lots where the natural statistical variation is pretty low across the lot. The space version of the same part is built in MUCH smaller quantities, so it the lot actually suffers worse statistical variation. So you need to do even more statistical lot screening to ensure your variability is acceptable, and of course that costs money. I’ve seen space parts (capacitors, diodes, etc.) were upward of 20% of the whole lot is set aside for screening.
whenever you lower the cost of something 100X you get an explosion of demand from industries which were not previously viable.
Logic closes on that. We all get that.
The issue is that nothing of any consequence in the space industry is coming down by 100X anytime soon. (This goes back to my earlier comment on discussions here often getting wires crossed on timelines, FTR.) Maybe in 3-4 decades launch might approach 100X reduction of a 2020's equivalent, but certainly not for the practical future.
SpaceX recently launched a program to build satellites for government purposes. I'm guessing that they intend to use their Starlink approach for those.. but that Space Force etc will want high reliability. Do you think they have a way to produce an up-market Starlink bus with Space Force Payloads for, say 6 times the price which would be $1.5M (and then charge the Space Force $5M for it)? Or do you think they are going to fall on their face?
Sorry for the overt diplomacy here...
Yes, it is reasonable to assume that SpaceForce! wants a satellite with higher reliability than what SX has done with Starlink. It is reasonable to assume SX could build that satellite, and it is also reasonable to assume they can charge SpaceForce! accordingly. There’s zero evidence SX is incapable of doing any part of building, launching, or operating such a constellation, and there's zero evidence to suggest SX would in any way fall short of whatever requirements SpaceForce! deems necessary.