What Improbable really has, or claims to have, is a way to scale massively multiplayer games where everyone is in the same world. They have a back end which is basically a synching system for small data elements between computers in a data center. Think of it as being like the way a shared memory multiprocessor synchs its memory system. Some data items belong to a single CPU, and that CPU has read/write access to it. Anybody else can read, but they get updated if the data changes. Who has write ownership can change based on demand. Improbable has a downloadable demo if you want to play with this.
The game use of this is as a solution to the "region crossing" problem in Second Life. The big Second Life world is divided into squares 256m on a side, each managed by a separate process. The processes talk to each other, and objects can move across region boundaries. The system used is ad-hoc and fails frequently. If you drive around Second Life, about twice an hour you'll have serious trouble at a region crossing and may have to relog.
Improbable raised $500M, the largest investment ever made in a European tech startup.[1] They put 150 developers on the problem. They claim to have a solution. It's set up as a pay per use service. You pay for data storage, you pay for network traffic, and it has to run on Google's cloud. They don't publish prices, but at least one developer dumped them for cost reasons.
This is going to be like physics engines. In the 1990s it was really hard, and physics engines required heavy research into nonlinear differential equation solvers and fast collision detection. I used to work on that. There were startups with unicorn dreams - Mathengine, Havok. Now, everybody does it, and multiple open source libraries are available. Mathengine is gone, and Havok had a down round, downsized, replaced the management, and was sold twice. Physics is routine middleware now.
Right now, though, Improbable has the only system for this. Until someone else figures it out. This event pushed Improbable to open source their API code with the MIT license. That means anyone can now see how to talk to the back end, and start thinking about how to replace their back end. Progress marches on.
(Hm. What if you wrote a back end which, instead of needing multiple "cloud" machines, ran on a single huge Amazon AWS instance, like the one with 128 CPUs. That's an easier problem than coordinating it over the network. You might not be able to support a planet sized world, but you could support large islands you have to teleport between.)
[1] https://www.theguardian.com/business/2019/jan/10/british-gam...