The Snapdragon 8 Gen 1 is a Samsung 4nm chip, and it might be the worst thing to happen to Samsung’s foundry in a decade. Qualcomm’s flagship for 2022 throttles hard, drains batteries, and runs hot enough that reviewers test it with thermal cameras, and the consensus forming across the industry is that the problem is not mostly Qualcomm’s design. It is the node. Samsung’s 4nm is not yielding the cool, efficient silicon a flagship needs, and this time everyone can see it.
4nm is meant to be a refinement of the 5nm process that already ran hot, the node where Samsung steadies the EUV line and fixes what the last generation got wrong. Instead it inherits the same disease and makes it worse. Yields are reportedly poor, well below where a mature node should sit, which means a lot of wasted wafers and a process struggling to hit its power and performance targets. When yields are bad the chips that do make it out are often the ones pushed hardest to hit spec, and pushed-hard silicon runs hot. The 8 Gen 1 is what that looks like in your hand.
Qualcomm did something almost unheard of, and it is the move that turns a bad node into a disaster. It took the refreshed version of the same chip, the 8+ Gen 1, and moved it to TSMC, onto a 4nm process in the same family that built the M1 so cleanly. Almost nobody re-fabs a flagship mid-generation. The result is night and day. The TSMC version runs cooler and lasts longer, the same architecture transformed by a better foundry. Qualcomm did not redesign the chip. It changed who built it, and the chip got dramatically better.
The entire pitch of Samsung’s foundry, going back to the Nvidia win on 8nm, was that Samsung is the credible alternative to TSMC, a little behind maybe but close enough and cheaper and available. The 8+ Gen 1 detonates that pitch in public. It is a controlled experiment, the same chip on two foundries, and Samsung loses it so badly the comparison becomes a meme. Every prospective customer now has a clean, quantified reason to doubt Samsung’s leading edge, handed to them by Qualcomm.
The damage runs past one chip and one customer to the thing a foundry actually sells, which is trust. A fabless company bets its entire product on the foundry’s process working, and Samsung just demonstrated, in benchmarks anyone can read, that its process might be the reason your flagship fails. Qualcomm will keep some business with Samsung for capacity and leverage, nobody wants to be a hostage to a single fab, but the default has flipped. TSMC is now the safe choice and Samsung is the risk, and that is a much harder hole to climb out of than a bad yield quarter. Samsung already runs in the mid-teens of foundry revenue against TSMC’s mid-fifties, and a public failure like this is how a gap like that stops being temporary.
Samsung’s answer, the one it is clearly betting on, is to skip ahead again. If you cannot win the current node on execution, win the next one on technology, and Samsung has something coming that TSMC does not yet have in production. Maybe that leapfrog works and 4nm becomes a bad memory. Perso, I doubt the next trick fixes this, because the problem is no longer which transistor Samsung has, it is whether anyone believes Samsung can build it well. The 8+ Gen 1 proved a Samsung chip can be fixed by taking it to Taiwan. That is the sentence Samsung’s foundry has to erase, and you do not erase a sentence like that with a press release about being first.