I saw the OnLive demo. It was very slick, and the concept seemed more than viable, it was inevitable.
> the company had deployed thousands of servers that were sitting unused, and only ever had 1,600 concurrent users of the service worldwide
They were burning through all of their money because they highly overestimated the audience. That's one of the main problems the cloud was made to solve.
Traditional GPGPU customers, and machine learners, usually use CUDA instead of OpenCL / VulkanCompute / DirectCompute. At least from my position is looks that way. I’m a freelance developer and my clients are picking CUDA in ~75% cases. The rest is DirectCompute, if computing on client PCs and hardware compatibility is required.
Furthermore, there are other technological advances that have come even in the past 2 years that help facilitate this, like Secure Reliable Transport protocol (https://www.srtalliance.org/).
Things were just too stacked against OnLive at the time to make it work, but most of those barriers have since been removed after they had gone under.
EDIT: https://www.pcgamer.com/gaikai-founder-blames-onlives-pc-foc...
I suspect that Google will have better infrastructure than OnLive, and that their customers will have better Internet.
It's not the edge cache itself that matters; it's that many ISPs peer directly with Google.
Google Cloud can route the audio/video/keyboard packets mostly over Google's private network and then only use the public internet once it gets to your ISP (or their transit provider). This provides Google with more control over how the packet gets to the end user.
Google provides a similar service to Google Cloud customers as the "Premium Network Service Tier".
It's the challenge of edge computing datacenter design.
However, a gaming session means to have a dedicated daemon running for you somewhere. I doubt this can be deployed anywhere on the spot, but perhaps they have some amazing technology for that.
CDNs aren't helpful if every stream is unique.
Unfortunately for me at least, what I want is more akin to a cloud VM where they can host my games and I stream them wherever, which this isn't it. I've already got games between Steam, uPlay, Origin, etc. I have zero interest in buying a game 2x just to be able to have more portable play.
It'll probably be a mix of pay-once games, lots-of-games-for-a-subscription, and pay-once, play-until-Google-shuts-the-service-down games.