Earnest Cavalli said:
Normally I'd dismiss this kind of thing immediately as being impossible, but when it comes to Perlman, I've learned not to make that mistake. I thought the idea of streaming playable Crysis 2 in high detail to a low-end PC via cable Internet was a total impossibility until OnLive proved me wrong, so if Mr. Perlman says he's broken the laws of physics in the quest for a better cell phone signal, I'm on board.
The streaming isn't the thing I have trouble believing in. It's the business plan I don't buy. OnLive really is a great example of the next bubble to me.
First, until recently, you just couldn't get GPUs cheaply (http://www.nvidia.com/object/gpu-cloud-computing-service.html changes the picture a bit). That means that, for better or worth, for each active user they'll have to have the equivalent of a gaming machine in their data centre. You can't just use off-the-shelf cloud GPUs, because you need high bandwidth/low latency interconnects between GPUs, CPUs and main memory to play games. If active users don't get their dedicated gaming rig, some specialized hardware is required that'd virtualize multiple gaming rigs.
Second, in order to get video across a limited internet connection in acceptable quality, they need to encode it in real-time with an aggressively compressing encoding profile. That sort of stuff requires more GPUs (doubtful) or dedicated hardware to the tune of some 10k per video channel (= active user).
Third, in order to get controller lag under control, they need to have the gaming rigs as close to the player (in terms of network location) as possible -- which means the ISPs' data centres. They've said they'd do that, but I can't think that ISPs will be willing to offer that cheaply.
The upshot is that I haven't the faintest idea how they plan to be profitable in the face of such costs. In all likelihood, that's not the plan, and they just intend to sell off quickly.
Given that, I'd place breakthroughs in wireless networking under the "hype building" category... but I'm waiting for the Ars Technica dissection of the tech
