The Snapdragon 888 in this year’s Android flagships is a Samsung 5nm chip, and the reviews keep using the same word: hot. So does the Exynos 2100 Samsung built alongside it for its own Galaxy S21. Two leading-edge 5nm parts, both running warmer and throttling sooner than anyone wants, both built by a Samsung foundry that was supposed to use 5nm to put the chipgate years behind it for good.
5nm is Samsung’s first full EUV node, the payoff for the bet it made at 7nm, where extreme ultraviolet does most of the patterning instead of cameoing on a layer or two. In theory this is where Samsung’s early EUV gamble compounds into a real lead, denser and more efficient than anything before it. In practice the chips coming off it run hot, which points at the thing customers fear most in a foundry: not whether the node exists, but whether it yields clean, efficient silicon at volume, and Samsung’s 5nm is not doing that as well as the spec sheet promised.
The contrast is brutal because it is sitting right next door. TSMC’s own 5nm launched a few months ago building the Apple A14 and the M1, and those chips are the talk of the industry for the opposite reason: stunningly efficient, cool and long-lasting, the best mobile silicon anyone has shipped. Same node name, two different planets. When a Samsung 5nm Snapdragon throttles and a TSMC 5nm Apple chip sips power, the node number stops meaning anything and the foundry’s name becomes the spec. That is exactly the comparison Samsung cannot afford to keep losing.
Samsung pushes nodes to market fast and aggressive, the way it filled its lines landing Nvidia on 8nm, and sometimes fast means shipping a process whose yields and power characteristics are not fully dialed in. That has always been the double edge. TSMC moves a beat slower and arrives with the node more mature. For years Samsung’s speed was its weapon. At 5nm the speed is starting to look like the reason its chips run hot.
Qualcomm is watching this closely, because Qualcomm is the customer with the most to lose. Its flagship Snapdragon runs hot in reviewers’ hands and in customers’ pockets, the blame lands partly on Qualcomm’s design and partly on Samsung’s process, and Qualcomm knows which part it can change. Part of why the 888 sits on Samsung at all is that TSMC’s 5nm was booked solid by Apple when the deal was struck, the same sold-out-leader dynamic that sent Nvidia to Samsung. Samsung came in with capacity and a price, the way the number-two foundry, stuck in the high teens of market share against TSMC’s more-than-half, has to. A company that has dual-sourced before, that remembers TSMC building its 835-era rivals cleanly, is not going to keep handing its crown jewel to a foundry whose node makes it look bad once TSMC has room again. The 5nm heat is not only a Samsung problem. It is a reason for Samsung’s biggest customer to start shopping.
Maybe 5nm matures. Nodes get better across their life, yields climb, later steppings run cooler, and there is a version of this where Samsung quietly fixes the process and the heat fades into a rough launch nobody remembers. Perso, I think the damage is more than technical. The story hardening around Samsung’s foundry is that its chips run hot, and stories like that outlive the yield curves that started them. Being first to EUV was supposed to make 5nm the payoff. Right now it mostly looks like a chip that runs too hot next to one that does not, and a customer doing the math.