Yep. I'm not overly impressed with the lack of details. His response to the impact of SSD is naive, suggesting the marginal improvements of PC SSDs is comparable with what we'll get next-gen in consoles designed around next-gen storage performance. I don't put much faith in it, just as I didn't the old XBone cloud power suggestions.
Stadia will be able to afford an amount of processing power per user based on subscriptions and perhaps bolstered by other services repurposing hardware on downtime. That hardware will be finite. I suppose the best case is you have a game with a point in it that requires 3x the base performance, and some servers that only server additional processing power in that part of the game. So, for 10,000 players, 10,000 units serving the game and another 20 sharing workloads for those players on demand at the points they are needed. As long as you have short spikes in workload, that sound feasible (although latency in spinning up workloads would be a notable hurdle to overcome), but the moment you are requiring a majority of time above that base unit cost, you are needing to employ more hardware per user at more cost.
The server advantage is far more justifiable in multiplayer games at which point being a streaming service in the cloud makes little difference. I guess streaming a game means not having to sync millions of bytes of game state data which is the problem with locally rendered cloud-enhanced games. Crackdown 3 streamed could have a lot more physics and persistence akin to the original demos. In that respect, streamed games genuinely could do more. You could have a 100 TF server simulating a game with 100 players in a match and steaming the video feeds (rendered by satellite servers) where the data set for local rendering would be impossible.