Starlink : Markets and Marketing

[ChrisWilson68]

I have seen articles and comments in online forums that phased array antennas are about $30,000 each and those are the new ones just introduced in the past year. The only entities that buy these things are airlines, the US military, government entities, etc. Many of the comments have been extremely doubtful that SpaceX and OneWeb are going to be able to get the prices down to consumer levels anytime soon.

I have no idea if this is true or not. I find it hard to believe that the parts involved cost anywhere near that amount. $30,000 probably takes into account low volume production, lots of R&D recovery, salesman commissions, etc.

But at the end of the day, I have no idea what a consumer quality phased array antenna is going to cost SpaceX on a per customer basis. I wish someone would provide us with some solid info.

This is of course a critical question.  The $30,000 cost for commercial vendors now seems plausible.  Here is an 8 channel chip for Ka band.  If you had a 16x16 array (256 elements) it would be about pizza box size.  It would require 32 of these chips for receive, and 32 of the corresponding chips for transmit, so 64 chips in all.  I found one source online for $136 each, so that's about $10K worth of chips.  This would appear to be the major expense, so a $30K consumer price seems about right.

Now the question is how much cheaper you could get by spending a bunch of money to build a custom chip that did exactly what you need and no more, and only at the exact frequencies you intend to use.  If, for example, you could put 64 channels on a chip and make it for $50, then suddenly you need only 8 chips (4 receive, 4 transmit) for only $400 chip cost per terminal.  Without doing detailed design, I suspect this may be possible (for example, knowing your frequencies in advance, you might use 8 reference signals at 45^o phase shifts, which could allow you to replace phase shifters with simple MUXs.  You can use tuned power amplifiers, etc.)

So overall, I think a $30K price now is plausible, but it might well be possible to bring the cost down to $1000 or thereabouts.   Of course the design of the chips and the masks to make them might well be about $10M-$20M up front.  So this would only make sense if you intend to make 10s of thousands of units, all for the exact same market.  This market does not exist yet, which is why no-one has done it already.  But LEO direct to many points might make this development practical.

The chip you pointed out is Silicon-Germanium, which is a much more expensive way to make chips than bulk Silicon CMOS, which most cheap chips are made of.  People use Silicon-Germanium because it's easier to do high-precision RF stuff in that process.

You might remember the legal battle between Broadcom and SpaceX a few years ago.  Broadcom was unhappy that SpaceX hired away a bunch of its engineers to work on Starlink after Broadcom proposed a solution for them.  Broadcom made its fortune by taking functions that people used to think you could only do in Silicon-Germanium and doing them in cheap bulk CMOS instead.  Not only is making the chips much, much cheaper in CMOS but you can also integrate the RF components directly with pure digital parts of the design, including processors, all on one chip.

SpaceX has been developing Starlink for years.  There's no way they didn't have plans for the customer premises equipment from the start, and there's no way they've been saying they're going to sell this to consumers if they had a plan that would cost $30,000 per ground station.

Silicon is going to be the long pole in developing the CPE for Starlink.  Doing the chips takes years from start-of-design to low-cost manufacturing.

SpaceX would not be spending hundreds of millions of dollars to launch Starlink today and saying they'll start service later this year unless they were already close to having the chips ready for CPE that costs them well under $1,000 per unit.  They might still be working on spinning the chips for manufacturability, they might be spinning the boards for manufacturability, they might be working on different industrial designs, but in terms of chips, I'm quite confident it's a solved problem.  Whether SpaceX designed them in-house or they have a partnership with an external company, expensive chips just aren't going to be a thing in the CPE for Starlink.

[DistantTemple]

If anyone is likely to put in the $ and R&D to design a custom chip for phased array antennas, is an Elon Musk company! Also He has proven track record in providing the interesting challenge and attracting the best talent. And of course seeing the process through, including chip production and successful implementation - in the HW3 FSD Tesla AI computer. ANd that is a whole new type of computer! a whole new architecture!
Packing 16, or 64 or whatever units of an existing architecture/system onto a chip appears to have less large leaps into the future. (Now, if AI was included to seamlessly switch between satellites faster than ever before .... no I won't go there... talking out of the proverbial...)
If it takes a few $100M's to develop, but the cost of the chip and board is <$100 for quantities of millions then he has cracked it. Will Jim Keller and Pete Bannon come and make yet another seismic change to computing, by designing this for SpaceX?
Cracking a cheap almost "throw away" phased array, looks like one of EM's classic "first principles" things.   Damn it, this is the guy that is designing chips to stick in peoples brains! If there is a non-zero chance of success, and the project really needs it to succeed, then throw everything at it until its solved!
Just read your comment:

Silicon is going to be the long pole in developing the CPE for Starlink.  Doing the chips takes years from start-of-design to low-cost manufacturing.

The FSD chip was started in 2016/2017 - and its been rolled out successfully since middle of 2019! so 3 years!

[su27k]

The chip you pointed out is Silicon-Germanium, which is a much more expensive way to make chips than bulk Silicon CMOS, which most cheap chips are made of.  People use Silicon-Germanium because it's easier to do high-precision RF stuff in that process.

You might remember the legal battle between Broadcom and SpaceX a few years ago.  Broadcom was unhappy that SpaceX hired away a bunch of its engineers to work on Starlink after Broadcom proposed a solution for them.  Broadcom made its fortune by taking functions that people used to think you could only do in Silicon-Germanium and doing them in cheap bulk CMOS instead.  Not only is making the chips much, much cheaper in CMOS but you can also integrate the RF components directly with pure digital parts of the design, including processors, all on one chip.

SpaceX has been developing Starlink for years.  There's no way they didn't have plans for the customer premises equipment from the start, and there's no way they've been saying they're going to sell this to consumers if they had a plan that would cost $30,000 per ground station.

Silicon is going to be the long pole in developing the CPE for Starlink.  Doing the chips takes years from start-of-design to low-cost manufacturing.

SpaceX would not be spending hundreds of millions of dollars to launch Starlink today and saying they'll start service later this year unless they were already close to having the chips ready for CPE that costs them well under $1,000 per unit.  They might still be working on spinning the chips for manufacturability, they might be spinning the boards for manufacturability, they might be working on different industrial designs, but in terms of chips, I'm quite confident it's a solved problem.  Whether SpaceX designed them in-house or they have a partnership with an external company, expensive chips just aren't going to be a thing in the CPE for Starlink.

They filed a patent for the phase array design in 2018: DISTRIBUTED PHASE SHIFTER ARRAY SYSTEM AND METHOD, not sure if it's for the satellite end or ground end, maybe both?

[ChrisWilson68]

They filed a patent for the phase array design in 2018: DISTRIBUTED PHASE SHIFTER ARRAY SYSTEM AND METHOD, not sure if it's for the satellite end or ground end, maybe both?

My guess would be that they use the same idea in both the satellite and ground equipment.  It's perfectly plausible that they might use the same chips in both, but it's also perfectly plausible that they might use different chips for the two.  If they use different chips, the two chips might re-use significant functional blocks.  For example, they might have more of certain functional blocks on the chips the satellites use and fewer for the ground equipment so that the satellite can handle more simultaneous signals at the same time.  That would mean the chips used on the satellite were larger and hence more expensive while those for the ground equipment were cheaper.  If so, they'd likely start off with the more-expensive chip first and use it for both the satellite and ground equipment prototypes at first, since that chip could do either job.  Then, they'd do a revised version for the ground equipment that cuts its cost by cutting out some of the functionality.

Also, at least initially, they could use the same die for both the satellite and ground sides but for the satellites only use those with all their functional units working.  For ground equipment, they could use those chips that have flaws that break some of their functional units and have fuses in the package that they can blow based on the testing to tell the software which units work and which don't.  Since they need so many more of the ground units, they'd still probably want to eventually have a smaller, dedicated version for the ground units so they can fit more dice per wafer, even though that would cut their yield.

[Eka]

SpaceX's chip(s) are done and in production if they are starting deployment en mass.

They filed a patent for the phase array design in 2018: DISTRIBUTED PHASE SHIFTER ARRAY SYSTEM AND METHOD, not sure if it's for the satellite end or ground end, maybe both?

This patent is for implementing one and two dimensional arrays of phased array phase shifting circuits on a single chip. It's part of what I've figured they needed to do. I was assuming a transceiver for each element, but this looks like it could do it with one transceiver for the whole array. That greatly reduces silicone area used. Sounds like the design has tuning elements for trace length differences on the antenna array PCB. That simplifies PCB design. I'm going to guess that they can put a consumer terminal on one chip, or at most 4 phase shifter chips plus a transceiver chip. Satellite and ground stations likely need more power and multiple phase direction offsets active at once so they use a multi chip solution designed to handle that. I'm also betting they split transmit and receive into different arrays to make it easier to handle multiple beams all transmitting and receiving at once on the same frequency. Note: There is a packaging reason to go with a multiple chip solution even if it fits on one chip. If the packaging costs go too high due to the number of IO pins needed, it may be cheaper overall to split it up and use cheaper packages.

So, I figure one chip for the consumer terminals, or at worst two with one repeated at most 4 times. For the satellites and ground stations I'm betting they are using separate transmit and receive chip sets. This is to handle the multiple beams needed. So there is a set of transmit or receive phase shifter array chips tied to either transmitter or receiver chips. The chip sets are tuned for ka or ku bands. As mentioned by ChrisWilson68, they made the satellite chip sets first, then the consumer terminal chip.

If people are wondering, I've worked on a couple hardware projects where custom chips were used. One was needing beyond beading edge speed and the other was for cost reduction. They both put a whole circuit board of chips onto one chip. I wrote both bit slice and regular CPU firmware for both.

[DistantTemple]

So Starlink sats are being deployed.... not just a few to test but now 300! and 120/month planned.
Also the military are successfully communicating through these sats, not only in a simple test, but now in practice battlefield trials.
Also SX are talking as if they can roll out some connectivity to N USA and S Canada, late this year.
Therefore they must have cracked this nut.
The military seems most convincing to me, for phased array success. For a fast jet travelling some 600mph, (possibly turning ?) to connect to satellite (s???) moving above at some 17,500mph and the system to work, means the signal direction is changing! QED.
OK the current system could be very expensive! But with 120 SATS a month heading for orbit, it would be a surprise if they plan to ditch this lot due to incomplete phased array work!
It remains possible, that the consumer receiver version is still (too) high cost, but even so I suggest the argument above indicates they have mastered the main tec hurdles.
If further improvements are needed for performance and cost - which I assume is likely, these will not stop customer roll-outs this autumn/winter!
It is another significant technological step for the world, that SpaceX has taken.
(Separately inter satellite links appear to have been shelved, as SX is talking about local ground stations. If the laser link hardware is not on these sats, that will have to wait for a future version.)

[whvholst]

Reggefiber and other much smaller hardware companies (there are at least 2 still very active in Overijsel only, and at least 10 overall in all provinces) were and are technically/financially part of the hardware part of KPN (KPNnetwerkNL now) from the day 1. It started somewhere in 2000-2001.

No, they were not. The take-over was approved in 2014, until then it operated independently. Yes, there was a financial stake in it, but no operational control. Which makes a huge difference.

Now back to the topic:
Nor the Netherlands, neither UK or France (there are some chances for Germany though  because of funny federal structure) will offer any support or "express" any basic interest toward Starlink. It's moot. Please stop it.
All these countries have extensive very high level state investments, impressive lobbying and closed "family like" guilds which make the chance of entering this market from outside next to impossible. What is even more important there is close relation between these countries in the IT communications  area.

This is an unnecessary broad brush. It is correct that the chances of government support of Starlink are zero. It is also correct that the French and German broadband markets can be considered failures of implementation of EU policies in terms of local loop unbundling. In contrast, the UK, the Nordics, the Netherlands and several Eastern European member states have quite succesfully liberalised their telecoms markets. And a lot of the barriers of entry in both France and Germany will be non-existent for Starlink. It will be hard to deny Starlink spectrum that is currently unused by any operator. Starlink is also a massive opportunity for any new mobile operators to roll out 4G and 5G networks in France and Germany since it allows any newcomer not to have to rely on leased lines of France Telecom or T-Mobile, the incumbents.

[ChrisWilson68]

(Separately inter satellite links appear to have been shelved, as SX is talking about local ground stations. If the laser link hardware is not on these sats, that will have to wait for a future version.)

The last I had heard the inter-satellite links were planned to start going up on satellites starting in 2021.  Unless there's been some update on that that I missed, I wouldn't use the word "shelved" but rather the word "delayed" -- perhaps that's what you mean by "wait for a future version".  My understanding is that the reason they're talking about ground stations is that they're starting commercial service this year, before the satellites with the lasers even start going up, and the satellites without the links will be part of the constellation for about five years, until 2025 or 2026.  So they need local ground stations for several years.

[wannamoonbase]

(Separately inter satellite links appear to have been shelved, as SX is talking about local ground stations. If the laser link hardware is not on these sats, that will have to wait for a future version.)

The last I had heard the inter-satellite links were planned to start going up on satellites starting in 2021.  Unless there's been some update on that that I missed, I wouldn't use the word "shelved" but rather the word "delayed" -- perhaps that's what you mean by "wait for a future version".  My understanding is that the reason they're talking about ground stations is that they're starting commercial service this year, before the satellites with the lasers even start going up, and the satellites without the links will be part of the constellation for about five years, until 2025 or 2026.  So they need local ground stations for several years.

It’s very Elon like to get something up and working and relentlessly iterate the design. SpaceX needs revenue to fund Starlink.   Starting US service, get market penetration and revenue while working on the V2 hardware.

I have no idea how hard the laser interlinked are, I assume crazy hard, but they must think it’s possible.   If they figure that out and start launching even larger batches with Starships, they’ll be global before people know what hits them. 

[ChrisWilson68]

(Separately inter satellite links appear to have been shelved, as SX is talking about local ground stations. If the laser link hardware is not on these sats, that will have to wait for a future version.)

The last I had heard the inter-satellite links were planned to start going up on satellites starting in 2021.  Unless there's been some update on that that I missed, I wouldn't use the word "shelved" but rather the word "delayed" -- perhaps that's what you mean by "wait for a future version".  My understanding is that the reason they're talking about ground stations is that they're starting commercial service this year, before the satellites with the lasers even start going up, and the satellites without the links will be part of the constellation for about five years, until 2025 or 2026.  So they need local ground stations for several years.

It’s very Elon like to get something up and working and relentlessly iterate the design. SpaceX needs revenue to fund Starlink.   Starting US service, get market penetration and revenue while working on the V2 hardware.

I have no idea how hard the laser interlinked are, I assume crazy hard, but they must think it’s possible.   If they figure that out and start launching even larger batches with Starships, they’ll be global before people know what hits them.

Yeah, this is a great example of Musk not letting perfect get in the way of good enough, while still continuing to work relentlessly on the better solution for the future.

They might never get the laser links working.  They might pivot to some kind of RF link between satellites.  They might abandon inter-satellite links entirely.  Or they might get the laser links working on schedule by the end of this year.

The beauty of their approach is that Starlink will very likely be successful under any of these scenarios.  If and when the inter-satellite links are working, it will be that much more successful, but they're not going to let someone else beat them to commercial service and miss the opportunity waiting for it.

[Luc]

I would like to speak to the idea that the E.U. Or nearby countries might “ban” Starlink somehow. I do not think they will, rather I believe they will view it as a revenue opportunity and find ways to lightly squeeze it to extract money. Once European competitors are in a position to compete, they may attempt to tilt the scales in favor of the home team, but that will run into political headwinds both at home and abroad.

I know part of the discussion is more hypothetical, speaking to whether or not they could, and for me that revolves around how ‘realistic’ it would be for them to do so politically.

[FutureSpaceTourist]

Published by Eric Ralph in NewsSpaceX
22 hours ago
SpaceX Starlink job posting signals serious interest in a growing multi-billion dollar market

A new SpaceX Starlink job posting hints that the company is very interested in an established multi-billion dollar market for high-quality satellite internet – a use-case its Starlink constellation should be a perfect fit for.

One of the biggest sources for a recent boom in global demand for satellite broadband services, in-flight connectivity (IFC) is a rapidly growing market well on its way to multi-billion dollar annual revenues within the next few years.

https://www.teslarati.com/spacex-starlink-job-posting-billion-dollar-market/

[ZachF]

A little comparison of capacity costs between Viasat, OneWeb, and Starlink:

-Viasat 2 has a throughput of 260GBps, and had a cost of $600 million, giving an installed capacity cost of ~$2,300,000/GBps.

-Viasat 3 satellites are expected to cost around $700m total per satellite, and have an impressive capacity of 1TBps. This gives an installed capacity cost of ~$700,000/GBps for the next gen Viasat 3.

-OneWeb satellites have a throughput of 8GBps and have a cost target of $500k per satellite, launched on Soyuz gives a per-satellite launch cost of $2.5m. Put all of these together and you have an installed of around ~$300,000/GBps. If you factor in the lower possible capacity factor from time over the ocean, this is not much better than what Viasat 3 should be able to do.

-Starlink satellites have an estimated launch + manufacturing cost of about $500k, with the launch costing more than the satellites. With a throughput of 20GBps, this gives and installed capacity cost of ~$25,000/GBps. Now, since the launch cost dominates the total cost, adding Starship to the picture should more than halve this. They are currently producing satellites at annual rate of around 1,500 per year... quadruple this (which is what you'd need to do to get a 30,000 satellite constellation), and I see no reason that Starlink's installed cost shouldn't drop below $10,000/GBps.

In short, $/GBps installed:
$2,300,000 Viasat 2
$700,000 Viasat 3
$300,000 OneWeb phase 1
$25,000 Starlink
$10,000 Starlink w/Starship

I don't really see how anyone else can compete. OneWeb could probably get to $100k-$150k/GBps with New Glenn, but that's still far beyond Starlink.

[ZachF]

Some more numbers from those cost findings:

Viasat will be launching roughly one sat a year in 2021-2023, this is $700m per year to add 1,000GBps of capacity annually.

OneWeb will be launching about ~320 satellites a year in 2020-21, this is ~$750m per year to add 2,500GBps of capacity annually.

SpaceX is launching ~1,500 satellites per year, which costs ~$750m for 30,000GBps of capacity.

In short all three are interestingly spending about the roughly same money for vastly different results...

...Now add Starship it, and would cost only ~$1.2b per year to launch 6,000 satellites with 120,000GBps of capacity! 

[RocketGoBoom]

https://www.thedrive.com/the-war-zone/32346/the-air-force-and-spacex-are-teaming-up-for-a-massive-live-fire-exercise

The Air Force And SpaceX Are Teaming Up For A 'Massive' Live Fire Exercise
The exercise will involve SpaceX Starlink satellites, a variety of military assets, and a new command and control system working together.

Just last week, Air Force acquisition chief William Roper told reporters at the Pentagon that the Air Force and SpaceX will conduct an event on April 8, together with other branches of the U.S. military, that will see SpaceX Starlink satellites link up with multiple armed forces systems in a “massive” live fire exercise.

Elon is actually building Skynet.

He tried to warn us about the dangers of AI, then he goes off and builds Skynet just to prove it to us .

The system goes on-line August 4th. Human decisions are removed from strategic defense. Skynet begins to learn at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, August 29th. In a panic, they try to pull the plug.

Sarah Connor: Skynet fights back.

[RocketGoBoom]

A little comparison of capacity costs between Viasat, OneWeb, and Starlink:

-Viasat 2 has a throughput of 260GBps, and had a cost of $600 million, giving an installed capacity cost of ~$2,300,000/GBps.

-Viasat 3 satellites are expected to cost around $700m total per satellite, and have an impressive capacity of 1TBps. This gives an installed capacity cost of ~$700,000/GBps for the next gen Viasat 3.

In short, $/GBps installed:
$2,300,000 Viasat 2
$700,000 Viasat 3
$300,000 OneWeb phase 1
$25,000 Starlink
$10,000 Starlink w/Starship

I don't really see how anyone else can compete. OneWeb could probably get to $100k-$150k/GBps with New Glenn, but that's still far beyond Starlink.

That all looks correct to me, except for one item. I don't see a launch cost included for Viasat 3.

From my research on that, Viasat 3 (version 3, and they have 3 satellites in production) has reserved 3 launches, one each with ULA Atlas 5, SpaceX Falcon Heavy and Ariane 6 heavy-lift variant with four strap-on boosters. Those are not cheap launch vehicles. How does that change your numbers above, besides the obvious that it makes Viasat less competitive?

[RocketGoBoom]

In short, $/GBps installed:
$2,300,000 Viasat 2
$700,000 Viasat 3
$300,000 OneWeb phase 1
$25,000 Starlink
$10,000 Starlink w/Starship

EchoStar HughesNet is building their next GEO satellite, for rural internet service, called Jupiter-3.

It has only half of Viasat 3 capacity at 500 GBps. The cost I found was $400 million, which doesn't include launch costs. So it seems a bit worse than Viasat 3.

[Robotbeat]

https://www.thedrive.com/the-war-zone/32346/the-air-force-and-spacex-are-teaming-up-for-a-massive-live-fire-exercise

The Air Force And SpaceX Are Teaming Up For A 'Massive' Live Fire Exercise
The exercise will involve SpaceX Starlink satellites, a variety of military assets, and a new command and control system working together.

Just last week, Air Force acquisition chief William Roper told reporters at the Pentagon that the Air Force and SpaceX will conduct an event on April 8, together with other branches of the U.S. military, that will see SpaceX Starlink satellites link up with multiple armed forces systems in a “massive” live fire exercise.

Elon is actually building Skynet.

He tried to warn us about the dangers of AI, then he goes off and builds Skynet just to prove it to us .

The system goes on-line August 4th. Human decisions are removed from strategic defense. Skynet begins to learn at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, August 29th. In a panic, they try to pull the plug.

Sarah Connor: Skynet fights back.

Starlink is no more "Skynet" than the regular commercial satellites the military already uses today.

Sorry.

[RocketGoBoom]

Starlink is no more "Skynet" than the regular commercial satellites the military already uses today.

Sorry.

Sorry, I should have included sarcasm emojis.

[ZachF]

From my research on that, Viasat 3 (version 3, and they have 3 satellites in production) has reserved 3 launches, one each with ULA Atlas 5, SpaceX Falcon Heavy and Ariane 6 heavy-lift variant with four strap-on boosters. Those are not cheap launch vehicles. How does that change your numbers above, besides the obvious that it makes Viasat less competitive?

https://spaceflightnow.com/2018/10/26/viasat-confirms-spacexs-falcon-heavy-will-launch-next-gen-broadband-satellite/

Viasat says it expects the cost of the first two ViaSat 3-class missions, including satellite, launch, insurance and payload expenses, is expected to be between $1.2 billion and $1.4 billion. The company plans the first two satellites to cover the Americas and the Europe, Middle East and Africa regions.

Because of delays and switching from FH, I picked the larger number... ~$1.4 billion for the first 2 of 3 Viasat satellites, and this includes launch costs and insurance, works out to ~$700m per satellite all in.