Last week, I predicted that much of the Internet and most cloud datacenters would launch into space in the next ten years. Today the only part of the Internet in space is a very small amount of “bent-pipe” access: signals which go from a user to a satellite and bounce back down to a ground station which feeds them into the terrestrial internet where all processing is done and all queries answered by internet-connected servers, many of them in cloud data centers. Responses follow a reverse path through a ground station, back to a satellite, and then to the user. Below is a possible roadmap to the orbital internet; reality will certainly vary.

1. Starlink and Iridium prove the practicality of internet access service based on Low Earth Orbit Satellites (LEOS)—done

See: The Internet and The Cloud Are Going into Space

2. Bent-pipe access via LEOS creates a huge market in orbit

On the supply side, Starlink will go from 1500 to at least 44,000 satellites. OneWeb, a European competitor, has launched 54 satellites and will go commercial this fall. Kuiper from Amazon will come online in the next couple of years. Starlink will get laser-based satellite-to-satellite working and be able to provide service in most of the world where it has no ground stations. On the demand side, all users of obsolete slow-responding geocentric satellites are converting to LEOS as fast as they can. Even though fiber currently provides faster access than satellites, fiber is slow to deploy and perhaps never will get to the end of the road. Fiber can’t provide mobile access, which is required both by people and the Internet of Things (IoT), which will soon include all cars. Iridium already provides mobile access, and so will Starlink soon. 5G is the main competitor for this market. For emergencies, no terrestrial solution is adequate for communication. Poles and towers are subject to the same catastrophes as the people who depend on them locally. Starlink says it has contracts for backhaul from remote cell towers not on the fiber network to the internet backbone. This is the first but won’t be the last example of cellular acting as a concentrator and distributor of traffic which passes through LEOS. Technology will increase the capability of LEOS service, and competition will drive down the price.

3. Caching for and in orbit

Caching in internet terms means storing replicas of frequently accessed information near the consumer of that information to sped response times and lower overall communication costs. Every time you click on a URL, a query goes to a domain name server (DNS) somewhere to look up the physical address on the internet of the website your query is headed for. For example, google.com converts to 8.8.8.8. At least part of the DNS directory will quickly be cached in all ground stations. Large ISPs often host their own domain name servers to increase responsiveness; Starlink will not be an exception. I’ll be astonished if Starlink doesn’t start caching DNS directories in access satellites shortly. Users will experience great responsiveness, and Starlink will save an exchange with a ground station for each truncated query. Companies like Akamai and Cloudflare operate content delivery networks (CDNs). On behalf of content owners, the CDNs cache copies of fairly static content (movies, for example, but also many other types of web pages) at locations around the internet. This is a form of hosting that saves content owners from owning huge data centers with huge pipes themselves and assures that the content is quickly accessible from all over the world that each content owner cares about. Whether Starlink will operate its own CDN or partner remains to be seen; what is certain is that terabytes of content will move into space to be “near” users of satellite access. At this point, we will see the first dedicated cache satellites. Access satellites will query them by laser.

4. Smart routing in orbit

Once satellites can talk to each other, they become routers and can manage the quality of service and optimize routing dynamically to some extent, just as terrestrial routers do. If a query can be answered in space by a cache satellite, the query will go there and get a very fast response with no bouncing around in the terrestrial internet. If a query does need to go to earth, it might be routed with a satellite hop or two to a ground station collocated with a data center that can process some or all the requests.

5. LEOS gain a speed advantage

With smart routing, caching, and content delivery hubs in space and at ground stations, a query sent through LEOS will often get a faster response than the same query sent through the terrestrial internet. All packets traverse a net of routers to get to their ultimate destination. Each router adds delay to the packet’s journey for processing and queuing time. Each satellite can get traffic to any other satellite with a maximum of four hops, usually less. If the packet is then served from a space-based cache or a data center with a download station collocated, which will be most data centers in a few years, there are fewer hops and more alternatives for dealing with congestion. The speed of light is also 50% faster in space than in fiber, but that is not as significant to response time as reducing the number of hops.

6. Peering in orbit

Once competitive networks are firmly established in space, satellites from Starlink, Kuiper, OneWeb and other operators will start to exchange packets with each other via laser; this kind of traffic exchange, which is called peering is already standard practice among terrestrial ISPs, even fierce competitors. They don’t do it to be nice; they peer because of Metcalfe’s Law: the value of a network is directly proportional to the square of the number of endpoints. The ISPs gain more by combining their networks than they do by keeping them separate. The same goes in space.

7. Terrestrial aggregation

People like high-frequency traders and very serious gamers, for whom every millisecond counts, will start to use LEOS for access even when they live in areas with fiber. Fiber operators will start to add routes directly to space from their networks. Mobile applications (think automated cars) which need rapid response will be connected mainly by cellular networks unless space technology has evolved quickly enough for them to connect to LEOS (which is possible). Even cell towers located on fiber backbone will start talking directly to LEOS to better serve their traffic.

8. Data centers in space—the cloud in orbit

Within five years (I usually underestimate time), major data centers will be in space for simple economic reasons. Data center location depends on where the traffic is and the local price of energy to run the data center, and its attendant air conditioning. Within five years, a high percentage of queries will be passing through space; solar power is free once you’ve repaid the capital cost of solar panels and launch; a/c isn’t needed in space. The physical bulk of a data center without a/c and built for zero gravity will be relatively easy to lift into orbit. Security concerns alone are enough to make governments and corporations want to replicate key command, control, and data out of the reach of terrestrial physical attacks. Amazon is the biggest operator of data centers on earth; they will move quickly to orbit; cloud providers who don’t offer an orbital location will be at a significant disadvantage.

9. Computers built for orbit

Currently, computers are built to run on earth. Their speed is limited by how fast electricity can travel through their circuits; the energy lost in transit becomes unwanted heat. A chip designed for use in space can be run at temperatures near absolute zero. At these temperatures, many materials become superconducting; they provide almost no electrical resistance. The computers in an orbital data center will be faster than their terrestrial predecessors. If there is still cyber-currency mining, it will be done in orbit where none of the energy used is polluting, and calculations can be done faster than anywhere else. Especially with stricter and stricter environmental controls, it will be hard to justify building another data center on earth!

10. New backbone, when needed, is built in orbit, not under the ocean

With space-based caching, orbiting computer centers, and traffic relayed through space from mobile sources and aggregators as well as individuals, there won’t be growing demand for terrestrial backbone. Just as the highway network disrupted the railroads because of greater routing and dispatching flexibility, orbital routing and processing will shrink the long-haul fiber demand. Even where communication is between two terrestrial locations, the shortest and cheapest route will usually be through space. Sure, a New York to London flight is a good way to get traffic between these two cities; but, if you’re going from Minneapolis to Birmingham, do you really want to connect twice in the hubs? Neither do your packets. Space is the realm of the most direct connections. This roadmap is just to demonstrate that there is a credible way for the internet and cloud computing to become mainly orbital. It surely won’t happen in exactly this way and may not happen at all. Since each step above lowers the cost of computing and communications, each of these steps—at least the ones which actually occur—will present enormous opportunities for innovation and entrepreneurs.