One does not simply walk into Mordor. As mentioned in the article, at the spec level it makes perfect sense. Examples include hot-swap processor and memory boards. But to actually go about designing large processor chips - that's not easy. It needs DNA, one simply can't put together such a team by ones and twos. Apple had to go buy PA Semi, a start-up that had serious expertise in chip design covering two processor families.
That apart, does anyone know how much capex and opex benefit a company like Google or Facebook will gain by designing their own chips, as opposed to using commodity silicon?