Starlink fundamental cost per GB equation (and comparison to competition)

[Robotbeat]

And Starlink is still in early Beta, and this is ignoring the massive latency advantage.

Exactly, you have to rate capacity as #users x average speed, not just the average speed. Being in Beta, the number of users for Starlink is relatively low. For instance, on that website, there were only 19 unique connections on the download side for Starlink on January 21st. There were 69 for Viasat and 446 for HughesNet. This is what you would expect given SpaceX has referenced thousands of invites being sent out repeatedly compared to the 1.6 million Hughes subscribers and .6 million US Viasat subscribers.

You are comparing a couple of 3 year+ old satellites being hammered with 100s of thousands of users compared to hundreds of  brand new satellites being hammers by thousands of users. In that regard, the 3x  advantage Starlink is seing in download over Viasat, isn't necessarily that impressive and could disappear under load. We will see, but the trend over time has been somewhat negative.

Do you have latency comparisons as well? Yes, I know that it's an unfair comparison, but it's just as useful to collect that data, as latency is also important, and often overlooked aspect of networking. Also, do you have data for uploads and not just downloads for comparison and tracking.

It's a COMPLETELY fair comparison, and according to the CTO of Cloudfare (big content delivery network provider), it's improving over time. https://forum.nasaspaceflight.com/index.php?topic=48297.msg2181535#msg2181535

[oldAtlas_Eguy]

And Starlink is still in early Beta, and this is ignoring the massive latency advantage.

Exactly, you have to rate capacity as #users x average speed, not just the average speed. Being in Beta, the number of users for Starlink is relatively low. For instance, on that website, there were only 19 unique connections on the download side for Starlink on January 21st. There were 69 for Viasat and 446 for HughesNet. This is what you would expect given SpaceX has referenced thousands of invites being sent out repeatedly compared to the 1.6 million Hughes subscribers and .6 million US Viasat subscribers.

You are comparing a couple of 3 year+ old satellites being hammered with 100s of thousands of users compared to hundreds of  brand new satellites being hammers by thousands of users. In that regard, the 3x  advantage Starlink is seing in download over Viasat, isn't necessarily that impressive and could disappear under load. We will see, but the trend over time has been somewhat negative.

Do you have latency comparisons as well? Yes, I know that it's an unfair comparison, but it's just as useful to collect that data, as latency is also important, and often overlooked aspect of networking. Also, do you have data for uploads and not just downloads for comparison and tracking.

It's a COMPLETELY fair comparison, and according to the CTO of Cloudfare (big content delivery network provider), it's improving over time. https://forum.nasaspaceflight.com/index.php?topic=48297.msg2181535#msg2181535

Every month an average of ~100 sats are added to the constellation. Right now that is ~ a 10% capability increase every month. This month 130. Next month Feb 180 or even possibly 240! So improvement is expected. But the real item here is that because the system is global with a global set of users the analysis has to be based on the global picture and not just the US picture when determining capability. As more landing licencing is given in more countries. The average of usage over a period measured in days of for any single sat can be reasonably calculated with a user density map for the Earth. SpaceX could do the calculation but doubtful that anyone else would have the complete database to be able to make the calculation. Sampling may work but is subject to accurate weighting calculations for each area of each county. Such that all calculations are no better than a guess that are definitely not accurate and may be very far off the actual.

[ncb1397]

And Starlink is still in early Beta, and this is ignoring the massive latency advantage.

Exactly, you have to rate capacity as #users x average speed, not just the average speed. Being in Beta, the number of users for Starlink is relatively low. For instance, on that website, there were only 19 unique connections on the download side for Starlink on January 21st. There were 69 for Viasat and 446 for HughesNet. This is what you would expect given SpaceX has referenced thousands of invites being sent out repeatedly compared to the 1.6 million Hughes subscribers and .6 million US Viasat subscribers.

You are comparing a couple of 3 year+ old satellites being hammered with 100s of thousands of users compared to hundreds of  brand new satellites being hammers by thousands of users. In that regard, the 3x  advantage Starlink is seing in download over Viasat, isn't necessarily that impressive and could disappear under load. We will see, but the trend over time has been somewhat negative.

Wow, you just don't quit, do you? I showed how you cherrypicked the data (conveniently skipping over the averages staring you right in the face)

You didn't discover anything. I described the situation including past dates.

Starlink still has a big advantage in upload by a factor of  ~7. And this is the first time that existing geostationary was beating starlink, who knows if that will continue or was a blip.

That is a direct statement that Starlink was winning at every single point with the exception of what I pointed out since the Beta started.

And for your information, the long-term trend is essentially flat for Starlink, neither increasing nor decreasing.

The best fit line for the website's average daily download speeds since the Beta started is negative. See below. I didn't say that it was "sharply negative". I described it as "somewhat negative". A factual statement with a proper adjective. Being confronted with simple facts should not engender a negative response. Maybe this is information that is not representative of what is actually happening in the wild? That is certainly possible, but as a description of the specific data made available here, it is where the evidence points.

[Robotbeat]

And Starlink is still in early Beta, and this is ignoring the massive latency advantage.

Exactly, you have to rate capacity as #users x average speed, not just the average speed. Being in Beta, the number of users for Starlink is relatively low. For instance, on that website, there were only 19 unique connections on the download side for Starlink on January 21st. There were 69 for Viasat and 446 for HughesNet. This is what you would expect given SpaceX has referenced thousands of invites being sent out repeatedly compared to the 1.6 million Hughes subscribers and .6 million US Viasat subscribers.

You are comparing a couple of 3 year+ old satellites being hammered with 100s of thousands of users compared to hundreds of  brand new satellites being hammers by thousands of users. In that regard, the 3x  advantage Starlink is seing in download over Viasat, isn't necessarily that impressive and could disappear under load. We will see, but the trend over time has been somewhat negative.

Wow, you just don't quit, do you? I showed how you cherrypicked the data (conveniently skipping over the averages staring you right in the face)

You didn't discover anything. I described the situation including past dates.

Starlink still has a big advantage in upload by a factor of  ~7. And this is the first time that existing geostationary was beating starlink, who knows if that will continue or was a blip.

That is a direct statement that Starlink was winning at every single point with the exception of what I pointed out since the Beta started.

And for your information, the long-term trend is essentially flat for Starlink, neither increasing nor decreasing.

The best fit line for the website's average daily download speeds since the Beta started is negative. See below. I didn't say that it was "sharply negative". I described it as "somewhat negative". A factual statement with a proper adjective. Being confronted with simple facts should not engender a negative response. Maybe this is information that is not representative of what is actually happening in the wild? That is certainly possible, but as a description of the specific data made available here, it is where the evidence points.

So again, you cherry-pick by not using the full available data. Speeds before your graph started were lower than they are now.

Don't hide behind "I'm showing simple facts" when you have been and continue to cherry-pick data to fit your narrative!

"the data is noisy because the service is early, so therefore I'll wait to post until the point in time when the random variations end up fitting my narrative" is cherrypicking. It's statistically the same thing as lying, bro. Quit gaslighting us.

I attached a graph of the *full* (not cherry-picked) data set the website you gave uses. A fitted line shows a positive slope. Essentially flat, but if you have to assign a slope to it, the slope is positive.

Just admit you were proven wrong and save some of your dignity.

[dlapine]

Latency data is not readily available from the site ncb1397 used, but they do include upload as well as download data.

The current 7 month averages (jun20 to Jan 21) from the testmynet site are, as of today (Jan 22 2021):

Starlink 40.6 Mbs down / 10.5 Mbs up
Viasat 12 Mbs down / 1.5 Mbs up
Hughes 17.6 Mbs down / 1.4 MBs up

Peak average measured speeds: (same time frame)
Starlink  98 Mbs down / 20.2 Mbs up
Viasat 56.2 Mbs down / 1.5 Mbs up
Hughes 31.1 Mbs down / 2 Mbs up

This as reported by the site directly is readily available.

Anyone have pricing specifics to go with this?

[dlapine]

That data does appear to be US only. 

It will be interesting to see what the averages look like by this July as that's a presumed 1500 sat operational time frame.

[ncb1397]

Anyone have pricing specifics to go with this?

ARPU for the last reported quarter for Viasat was slightly over $100. Hughes is like $98.50. Starlink is $99 in the US, the equivalent of $101.26 in Canada and $114.93 in the UK.

[ZachF]

As a comparison:
Viasat-3 is each $650m, is in GSO so it has 100% geographical capacity factor (BUT still has time of use capacity factor, maybe around 50%?), is 1Terabit/s in capacity and 15 year lifespan (but whether that’s usefully different than, say, 7-10 years depends on how fast cost of data transfer goes down).
So that’s about 1-4 cents per GB depending on how you count.

The latency is terrible, though! That might mean they have a low useful capacity factor because few people want to use it.

Interesting thread.

Something to think about...The cost declines in data transfer front loads the total income a sat could generate to its first years.

The growth in satellite throughput per kg seems to be in the area of 15-18 percent per year, so the value of a gigabit of capacity likely declines by a similar amount annually.  Back when ipstar6/Thaicom-4 launched ~15yrs ago its ~40GBPs was a significant  portion of world satellite throughput, now it's equaled by two 260kg Starlink sats, or ~1/50th of Viasat 3.

If the value of throughput declines by 15 percent per year, that means that at the end of a 15 year lifespan for a satellite like Viasat 3, the value of its capacity will have dropped by a factor of 10.

A 15 percent annual decline also puts the total life income of a 15 year satellite at the equivalent of 6 years of worth of first year income.

(edited for spelling/grammar, that's what I get for using my phone...)

[matthewkantar]

As a comparison:
Viasat-3 is each $650m, is in GSO so it has 100% geographical capacity factor (BUT still has time of use capacity factor, maybe around 50%?), is 1Terabit/s in capacity and 15 year lifespan (but whether that’s usefully different than, say, 7-10 years depends on how fast cost of data transfer goes down).
So that’s about 1-4 cents per GB depending on how you count.

The latency is terrible, though! That might mean they have a low useful capacity factor because few people want to use it.

Interesting thread.

Something to think about...The cost declines in data transfer front loads the total income a sat could generate to its first years.

The growth in satellite throughput per kg seems to be in the area of 15-18 percent per year, so the value of a gigabit of capacity likely declines by a similar amount annually.  Back when ipstar6/Thaicom launched ~15yrs ago its 40GBPs was a significant  portion of world satellite throughput,  now it equaled by two 260kg Starlink sats, or 1/50th of Viasat 3.

If the value of throughput declines by 15 percent per year, that means that at the end of a 15 year lifespan for a satellite like Viasat 3, the value of its capacity will has dropped by a factor of 10.

A 15 percent decline also puts the total life income of a 15 year satellite at the equivalent of 6 years of first year income.

Launch costs are dropping too, though in a choppier manner.

SpaceX didn't come up with five year life spans for the on orbit gear out of thin air. I am sure they ran the numbers and found the sweet spot.

[ncb1397]

Don't hide behind "I'm showing simple facts" when you have been and continue to cherry-pick data to fit your narrative!

"the data is noisy because the service is early, so therefore I'll wait to post until the point in time when the random variations end up fitting my narrative" is cherrypicking. It's statistically the same thing as lying, bro. Quit gaslighting us.

I attached a graph of the *full* (not cherry-picked) data set the website you gave uses. A fitted line shows a positive slope. Essentially flat, but if you have to assign a slope to it, the slope is positive.

Just admit you were proven wrong and save some of your dignity.

Your chart is sort of weird. X-axis is year(but goes back to 2017). Data points also are evenly spaced when early data was sparse  (many days had no reported average daily download speeds). Anyways, the trends are bifurcated into two distinct phases of deployment. To illustrate....

[Robotbeat]

Don't hide behind "I'm showing simple facts" when you have been and continue to cherry-pick data to fit your narrative!

"the data is noisy because the service is early, so therefore I'll wait to post until the point in time when the random variations end up fitting my narrative" is cherrypicking. It's statistically the same thing as lying, bro. Quit gaslighting us.

I attached a graph of the *full* (not cherry-picked) data set the website you gave uses. A fitted line shows a positive slope. Essentially flat, but if you have to assign a slope to it, the slope is positive.

Just admit you were proven wrong and save some of your dignity.

Your chart is sort of weird. X-axis is year(but goes back to 2017). Data points also are evenly spaced when early data was sparse  (many days had no reported average daily download speeds). Anyways, the trends are bifurcated into two distinct phases of deployment. To illustrate....

Whatever you want to use to justify your cherry-picking. I just extracted the data from the graph image and plotted it within any massaging.

[ncb1397]

Don't hide behind "I'm showing simple facts" when you have been and continue to cherry-pick data to fit your narrative!

"the data is noisy because the service is early, so therefore I'll wait to post until the point in time when the random variations end up fitting my narrative" is cherrypicking. It's statistically the same thing as lying, bro. Quit gaslighting us.

I attached a graph of the *full* (not cherry-picked) data set the website you gave uses. A fitted line shows a positive slope. Essentially flat, but if you have to assign a slope to it, the slope is positive.

Just admit you were proven wrong and save some of your dignity.

Your chart is sort of weird. X-axis is year(but goes back to 2017). Data points also are evenly spaced when early data was sparse  (many days had no reported average daily download speeds). Anyways, the trends are bifurcated into two distinct phases of deployment. To illustrate....

Whatever you want to use to justify your cherry-picking. I just extracted the data from the graph image and plotted it within any massaging.

All it is is applying the principles of time-series segmentation.

It is often the case that a time-series can be represented as a sequence of discrete segments of finite length. For example, the trajectory of a stock market could be partitioned into regions that lie in between important world events...

https://en.wikipedia.org/wiki/Time-series_segmentation

It is equivalent to plotting global average temperatures and segmenting the data based on the industrial revolution (~1750-1800). If pre-industrial temperatures were decreasing, that doesn't mean the current trendline is flat.

[Robotbeat]

“Time series segmentation” is another way of cherrypicking. Pick the right segment with noisy data and you can make it say whatever you want.

[oldAtlas_Eguy]

Speed tests are a little bit disingenuous. Because they test the speed between you and the speed test server. This server is at some datacenter somewhere as is definitely not the exact same physical server every time the test is run. Differences in the data load paths within the datacenter as well as the total throughput load the data center is seeing can skew results. Also the data load on the Internet backbones between the speed test datacenter and the Starlink Gateway also may weigh in on lowering sometimes the values. As is also there is no guarantee with SpaceX putting in more Gateways for any speed test run at any one time uses the same Gateway.

And then along with all of this is the fact that the Starlink network is constantly changing with more subscribers more sats and more gateways. Any trends seen in the speed test results may be an artifact of any one of the many items that can affect the data speed between any specific user and any specific datacenter as well as the the specific hardware server within the datacenter. It is about as close to random as can be achieved. What is to be looked for is large magnitude changes that persist over significant amount of time a year or more of a factor of 2 or more. This would then show whether SpaceX is ahead or behind the curve on supply vs demand for data by its subscribers. A near flat average shows everything is proceeding exceedingly well.

ADDED: I hate auto spell checker correction.

[Robotbeat]

Yup.

If you just look at the data, the spread is all over the place. I don't think it's reasonable to say with confidence it's going up or down in any real sense.

If you want to be critical of Starlink, point out the large spread of speeds in the graph (i.e. that speeds change a lot between tries), not a miniscule slope (which is positive, BTW) in what looks like almost totally random data (because even with truly random data with a flat distribution, there will be some slope until your sample size approaches infinitiy).

[macpacheco]

I work in telecom, I work for small ISPs and PSTN companies.
We can't come up with a reasonable cost per GB because not even SpaceX knows the right assumptions to come up with such a number.
The reality is datacom / internet is a high upfront investment and high ongoing margins segment.
One of my customers has thousands of miles of fiber rolled, including around 300 miles of a state wide fiber ring.

Back to Starlink. Let's make a mental exercise to understand how complicated it is:
the world has 149 million km² of land.
Let's say Starlink ends up with exactly one customer per 10 km² of land, spread out evenly across the world.
around 10 million users. Each paying US$ 100/mo. US$ 1 billion/mo revenue.
10m users / 30k satellites = 333 users / satellite
But as most of the world's surface is water, let's make it 1000 users / satellite.
Can Starlink handle 1000 users/satellite with 100Mbps broadband service ? Let's assume an average of 10Mbps of bandwidth demand per user, or 10Gbps per satellite.
Sounds very reasonable. Except in the real world there will be very few customers in the middle of the Sahara/Amazon rain forest. And lots of customers in rural north America / Europe.
Perhaps 10Mbps / user average is too much.
If users clump too much there might be 100Gbps of transfer demand on some satellites in congested areas and next to zero in remote ones.
In essence that's how one would model Starlink's business.
The benefit is such a model might work even with limited ISL capability. My assumption is there will eventually be a Starlink V2 satellite that will be bigger/heavier/far more powerful that will only be launched with Starship. Say still 60 satellites per launch but 1 ton satelites (4x heavier than now). Perhaps 4x the solar panel power to handle lots of laser links and a beefier RF capability.
From that standpoint the current constellation is just a major beta compared to what it will be.
But if you follow my numbers you will understand why it won't be possible for Starlink to actually compete with comcast/warner cable anywhere they have fiber or even where its copper only but they have faster than say 50Mbps service. I mean actually hurt their business. Starlink could steal a few percentage of the customers, but it would quickly saturate the satellite coverage in the area.
Let's think about the extended Boston area. If you are inside the route 128 ring, there is fiber. Starlink pretty much can't compete there. Once outside the 128 ring, fiber coverage is spotty, I have a few friends/former co-workers that complain (a lot) that they can't get 100Mbps service there, at all.
Starlink has the potential to make inroads there. But I should assume they can't service more than around 10k customers in the area. An area that has several million people that has internet service available, but might have good use for 100Mbps speeds. It is a very high tech area, lots of employees actually VPN to work full time now. The company I used to work for (half a billion/year in sales, a company that competes with Oracle/MySQL/...) just had a company wide pool where 70+% of the employees would prefer to stay full time telecommute even after the pandemic is gone. I assume that profile is the norm.
However it could be possible that just the people that live in the rural zone of Mass/NH/CT/RI/ME might be 50k people.
There will be clumps of high demand area all over North America/Europe.
I see starlink having a few million users already as a sure thing, even with a US$ 100/mo price tag and US$ 500 install cost.

Of course all of that is just an educated guesstimate analysis. Mostly just thinking out loud, although I'm in the business.
But keep one thing in mind. SpaceX must have around 50% margins on gross revenue vs fixed monthly costs. Even considering the cost to keep refreshing the constellation over time. No sane telecom company would work for less.

Oh and forget about Starlink broadcasting TV. That's not what the system was designed to do. That's what GEO satellites are perfect at. 2 second TV delay isn't a big deal, mostly impossible to perceive. Of course lots and lots and lots of Starlink customers will stream video, including realtime TV, but TV service shouldn't be a primary announced capability as that would reduce how many customers/area they can serve.

I wouldn't be surprised if it turns out 10Mbps / user for 100Mbps service is too much. Its very typical with very high speed internet for 50:1 over subscription ratios. Internet consumption tends to be bursty. I have 300/300 GPON service, in the start I downloaded up to 2TB/month, but then it settled down to a point where that 300Mbps is mostly a way to download faster, but 100Mbps would serve me equally as well.

Just some food for thought by someone that actually works with Internet all day long.
Oh, one of my non ISP customers just obtained cheap broadband 600/300Mbps for US$ 50/mo in a city with <200k people. Brazil's internet landscape is much, much, much more competitive than USA, we just don't have that ability of the big guys suing the little guys to oblivion. I have 6 fiber internet alternatives right where I live. And most ISPs already offer affordable service with at least 100/50Mbps speeds, some reaching 300/150Mbps speeds.

This was enabled by a pool of little ISPs that founded a backhaul coop Forte Telecom. They offering 100Gbps back hauls from here to São Paulo (1100km away the main internet hub of South America). Something like that was simply unthinkable.
The ISP I work for just activated (under a week ago) a 100Gbps back haul with traditional carrier.

Even with all of that, I'm certain Brazil has at least a million people in the rural area that has no access to land based internet that can afford US$ 100/month and need quality internet with at least 50Mbps, mostly owners of rural property.

So don't obsess about Starlink offering service in sub urban areas. That's not going to be the bread and butter. It will be rural/edge of sub urban areas. Keeping the monthly fee high enough will be required to avoid a demand level they might not be able to handle until V2 starlink is broadly built out.

Okay, enough rambling. 

[Robotbeat]

It helps not to have to dig cables in the ground, even in suburban areas. So even if your backhaul is cheaper with fiber (which depends on a lot of assumptions that may not last too long), your last mile will remain fairly expensive with cable in the suburbs.

And Starlink can compete just by buying a hungrier business. They overcome the inherent cost difference by operating a lot more lean. Comcast isn’t known for its efficiency.

In the urban core with multistory buildings, though, I kind of agree that Starlink won’t be more than a niche competitor.

[Nomadd]

Comcast isn’t known for its efficiency.

That made me laugh. Comcast is known for spending about 20% of their revenue (and that's being charitable) on the system and blowing the rest on acquisitions, stupid investments and hiring everybody they know for rarely show jobs. If they'd reinvested their incoming cash into developing and expanding the system, they could have made the old Ma Bell look like minor league players.

[r8ix]

In the urban core with multistory buildings, though, I kind of agree that Starlink won’t be more than a niche competitor.

We don't (yet) have real competition among ISPs in the US, generally, but my last apartment in central Russia (2017) I had 120Mbs symmetrical fiber for about $7/month ($3 in June, July, and August). I also had at least 17 providers I could choose from…
Similar situation with cell service, too.
Most of the US is served by function monopolies or duopolies when it comes to internet service, and Starlink will be very welcome as an additional option.

[oldAtlas_Eguy]

In the urban core with multistory buildings, though, I kind of agree that Starlink won’t be more than a niche competitor.

We don't (yet) have real competition among ISPs in the US, generally, but my last apartment in central Russia (2017) I had 120Mbs symmetrical fiber for about $7/month ($3 in June, July, and August). I also had at least 17 providers I could choose from…
Similar situation with cell service, too.
Most of the US is served by function monopolies or duopolies when it comes to internet service, and Starlink will be very welcome as an additional option.

An addition to the lack of ISP competition is the traditional control over permits and right of way leases in the US. These are controlled by the Local government. A lot of times a provider obtains a exclusive lease for Telephone or Cable TV service to be strung or buried in the right of ways. Either of the 2 lease holders can provide Internet. As in my location the Telephone provider AT&T and the Cable TV provider Mediacom. It is mostly a case of existing long term leases and a slow continuous technology evolution of both from wired transmission to fiber transmission so that now both offer 1Gbps Internet. For another provider to come into the are they would have to somehow obtain the permits and leases as well as string the fiber/cable. The permit and long term leasing processes are slow and are a major road block for competitors to put in service in an area. Basically for a long time it has been effectively local government controlled monopolies for telephone or Cable TV services. But such has not long ago disappeared and what remains is high up front cost barrier to other providers from offering service. It is mainly because in the US the providers own all of their infrastructure they use to provide service instead of leasing a data capability over a state owned infrastructure. In the second provider model to be a ISP provider just requires the connection to the state owned infrastructure and a connection to the Internet backbone by your data center with your DNS and etc servers exist.