Unlike Bezos and Branson, they’re going to stay there.

Today we have space-based internet access and a terrestrial internet; within ten years, we’ll have a space-based internet. Internet traffic will travel more miles in space than on terrestrial fiber. By that time, the great cloud data centers of Google, Amazon, Microsoft, and their competitors and successors will mostly be in orbit as well. Five years from now, this transition will be obvious, accepted, and well underway—or this will turn out to be the dumbest prediction I’ve ever made. Starlink is not the cause of the Internet moving to space; it’s an early example of the technologies which are enabling the move.

The primer (nerds feel free to skip)

Today, the components of space-based internet access are a user radio and an antenna, usually but not always dish-shaped, satellites, and ground stations with their own radios and dish antennas that provide access to the terrestrial internet backbone. Techies like to call the satellite link a “bent-pipe.” Your packets go through your radio to a satellite; they are sent back down to a ground station; they go to whatever their destination is on the Internet; a reply comes back to a ground station, goes back up to a satellite, and finally back down to your radio and you. Note that even the simplest query takes four hops through space, and so the time space hops take is very important to how long it takes for you to get a reply. We call the time between query and reply “latency.” The time a space hop takes depends mostly on how far away the satellite is since radio signals travel at a constant 186,000 miles/second in a vacuum.

The first satellites used for internet access almost 20 years ago were geostationary (GEOS): 24 hours/day, 365.25 days/year. They are at the same spot in the sky as seen from any spot on earth. Having them stationary means it’s possible to aim a dish antenna at them; a directional antenna like a dish is an energy-efficient way to communicate point-to-point. One big problem, though: the laws of orbital mechanics say that GEOS must be 22,000 miles high. Four hops of 22,000 miles each require more than 0.5 seconds to complete, even ignoring any delay in the satellite and on the ground. The human ear is sensitive to delays of more than 0.15 seconds. Zoom-like interactivity is essentially impossible although streaming, since it’s one way, works fine. Even web browsing is very painful using GEOS because the typical web page takes many interactions to assemble itself.

Companies like WildBlue offered internet access using GEOS. HughesNet and Viasat still do in the US. Bandwidths have gone up since I wrote about my experience with WildBlue (Why Satellite Internet Access Sucks) but the latency problem with GEOS is incurable, and the bandwidth of the technology they use makes the companies impose data caps which are unrealistically low by today’s standards. Users of GEOS are eager to convert to almost anything else (except possibly DSL), but they don’t usually have other options available where they live.

The solution to the latency problem is low earth orbit satellites (LEOS). These orbit only a few hundred miles above the earth, so the travel time of each hop is negligible. However, being low, they orbit the earth in about 90 minutes and appear to zip across the sky. There’s no way to aim a dish at them mechanically. With a small non-directional antenna, the amount of data that can be exchanged without expending a huge amount of energy and frying objects in the immediate vicinity is small.

Nevertheless, Iridium, which has gone through bankruptcy once, has several very useful mobile applications based on a fleet of 66 LEOS in polar orbit which covers every spot on earth. They’ve improved since I wrote about them (Going Sailing) and were essential to a sail I took from Norway to Scotland and back.  Their LEOS, most of which were launched under contract by SpaceX—parent of Starlink, can communicate with each other, so a packet may take several hops in space before finding a satellite that can see a ground station. The advantage of this approach is that they only need four ground stations, and the service works even in countries that don’t allow ground stations to be built.

Only one of Iridium’s services is internet access, strictly speaking. It is available with or without a fairly expensive but usable voice service. The bandwidth is low—think dialup speeds, but when you’re in the middle of an ocean and need a weather forecast or other info, it’s far better than nothing. Iridium is packaged with products like Garmin’s InReach, which leaves breadcrumbs on a website wherever you roam and can also be used for an SOS anywhere the sky is visible. A large number of devices with low but urgent data needs (the Internet of Things) take advantage of Iridium, as do aircraft. The Iridium access devices have small, non-directional antennas which can connect with the LEOS, which zip by. Because the antennas are non-directional and battery-operated radios are low-powered, Iridium is not suitable for high-bandwidth applications.

And now there’s Starlink with some 10,000 users as of February (a very small number), perhaps 500,000 people who’ve put down deposits and are waiting an indeterminate amount of time for service, still in beta but a proof of concept for delivering affordable high bandwidth to remote places with LEOS. Starlink antennas are dishes, but they track satellites without moving the dish using an electronically shaped beam—an important breakthrough for high bandwidth. SpaceX has pioneered reusable rockets, so they can launch swarms of satellites cheaply—about 1500 are in service now, with 120 or so new ones launched each month.

Starlink has an agreement to collocate ground stations with Microsoft and Google cloud datacenters—a development not lost on Amazon, which has received licenses to launch its own competitive satellite constellation called Kuiper and has started a space division.