Analyst Probes Speed Claims of Starlink’s Broadband Satellites

Financial analyst firm Cowen has attempted to predict the capability of SpaceX’s new mega constellation of small Low Earth Orbit (LEO) Starlink satellites, which they claim should eventually be able to serve 485,000 simultaneous data streams in the USA with 100Mbps speeds or 1.5 million streams with over-subscription.

SpaceX will soon – after the next launch – have 773 LEOs in orbit (each weigh 260kg) and they recently began a closed beta trial in North America, with an open beta expected in November 2020 (limited commercial services to follow). As it stands the satellite operator has an initial ambition to deploy 4,425 LEOs by 2024, which could potentially be followed by up to 12,000 at a later date (possibly late 2026).

Earlier this month the operator claimed (here) that beta testers were seeing download speeds “greater than 100Mbps” (Megabits per second) and latency times of 18-19ms (milliseconds), although they added that “round trip to internet” latency times of faster than 40-50ms may be closer to the real-world experience.

Admittedly this might not be quite akin to the “gigabit” or “fibre-like” levels of performance that have been stated in the past, but it’s still good progress. Such speeds will also still be very attractive to those stuck in rural or poorly connected areas, which is a key market for Starlink.

New Starlink Research

However, analysts at Cowen point out that Starlink’s submission to the FCC noted that “each satellite in the SpaceX system provides aggregate downlink capacity to users ranging from 17 to 23Gbit/s” (1Gbps = 1000Mbps+). Assuming the unrealistic position of 100% efficiency, and a top capability of 20Gbps per satellite, Cowen said that each LEO could thus handle 200 simultaneous streams (users) at 100Mbps.

In other words, when looking at the situation across just the USA, and only if all 12,000 LEOs were deployed, the firm predicts that Starlink could serve 485,000 simultaneous data streams at 100Mbps. The caveat here is that residential connections share their capacity in order to keep costs down and most of us aren’t saturating the max speed of our connection 100% of the time, which means you can cater for more users than this.

Cowen Statement (Lightreading):

“Equating this number to actual customers depends on Starlink’s policy on oversubscription rate, for example assuming a 3x oversubscription rate could imply a 1.5 million total addressable market (not all customers will be online simultaneously, far from it, though we believe the post-COVID era will drive more need for internet reliability, thus more conservative oversubscription levels).”

The more you share out that capacity, the greater the number of users that you can serve, but also the greater the risk that network congestion will creep in to suppress the speeds for some of your customers. Congestion is one of the reasons why more established satellite operators often suffer big performance drops during peak times.

We should point out that Cowen’s formula is quite simplistic and doesn’t appear to factor in the impact of ground stations, caching, peering / routing, package choice (slower tiers?) and various other factors (e.g. cost). Indeed, we suspect that, privately, SpaceX won’t yet have figured out precisely how best to balance their network (the beta will help), but we wouldn’t be surprised if they adopt a far greater level of “oversubscription” than above.

However, the key point Cowen makes is that Starlink does have clear limitations and so it won’t be able to solve all of the connectivity problems that exist. “Our analysis shows that LEO satellites will continuously be a step behind wireline telco/cable operators in meeting US consumer demand for broadband,” said the firm. In fairness SpaceX’s boss, Elon Musk, has already said that “Starlink is not some huge threat to telcos.”

All of this discussion is of course somewhat academic given that the most important thing ISPreview readers will want to know is when can they get it in the UK? On that point, SpaceX doesn’t yet have an answer and they’d probably need a few ground stations first.