IMO, this was one of the more interesting releases at re:invent this year. It seems AWS is really making a move to deploy AWS outside of the Region.

AWS Outposts let you have AWS on-prem, AWS Local Zones puts the AWS data plane closer to users and now AWS Wavelength brings AWS services to the edge of the 5G network.

Although I can't personally conceive of all the use cases, I can definitely imagine there are a million different applications that could benefit from being deployed that close to customers.

This feels like another positive sign that the de/centralized pendulum is about to swing back, as the centralizers are running out of momentum.

If low latency were actually in demand, this would be reflected in our popular runtimes and frameworks. As it stands, opening phone apps can take seconds, embedded devices shamelessly have hundreds of milliseconds touchscreen lag, and commercial webpages spend much time loading a host of malware and backhauling surveillance data.

Users do enjoy low latency, just not with enough importance to prevent the market from adding bloat up to what they're willing to tolerate. If this innovation just results in 10-20ms to load more surveillance scripts, that's not particularly noteworthy!

While I am all for the decentralisation trend where it matters, I think you miss some interesting use cases for this, like running compute intensive software on the server instead of on the client. Latency often matters in use cases like this, as evidenced by the recent Stadia threads.

One of the early users they mention is the Finnish VR hardware company Varjo - imagine being able to have a monster GPU on the server doing all the hard work, and having a lightweight non-tethered headset

For compute intensive applications, doesn't compute time far exceed the network latency?

If compute time is what makes up the bulk of the response time, moving the server to the edge won't really reduce the response time significantly.

Consider drone autopilot. Probably too heavy to run AI onboard. But latency needs probably single digit ms. So pop in a 5G card and this would solve that problem. (Though what to do when connection blacks out is a completely different problem).

ah, I meant compute heavy but realtime CPU/GPU tasks, which you can't run on a mobile client.

What mobile devices are capable of is getting pushed further each year, but photorealistic untethered VR for instance you still can't run on a mobile CPU/GPU, and you can't strap a top of a line Nvidia GPU and a large PSU to your head