Second, the sheer costs of building newer nodes have made those early costs less attractive to companies like AMD and Nvidia as well. This had its own knock-on effect, where the foundries now pursue secondary nodes that are less attractive to high-power manufacturers in the first place. We’re already hearing rumors that 10nm, like 20nm, may be a short-lived node for some companies, and it wouldn’t surprise us if AMD and Nvidia opt for second-generation 14nm hardware and skip 10nm altogether. This could be somewhat complicated if one foundry focuses on 10nm and one does not (Samsung and TSMC appear to be pursuing somewhat different strategies here, while GlobalFoundries is skipping 10nm altogether and heading straight for 7nm). But whichever nodes shake out as the long-lived nodes at TSMC, Samsung, and GlobalFoundries, it now makes more sense for many customers to follow the adoption curve rather than trying to lead it. For the foundries, the opposite is true — whichever company can offer a node first tends to capture a significant amount of revenue from that node. As always, keep in mind that the foundry nodes are also different between Intel and the pure-play foundries — what Samsung and TSMC call 10nm isn’t going to correspond to what Intel refers to as 10nm.