Samsung’s 2nm process, the one it calls SF2, is ramping now, and the company is staking its foundry comeback on it. The Exynos 2600 built on Samsung 2nm is meant to go into the next Galaxy flagship and prove, finally, that a Samsung leading-edge chip can run cool and yield clean. After the 4nm humiliation and the 3nm that shipped first and sold to almost no one, this is the node that has to change the story, and Samsung knows it.
Samsung has been building gate-all-around transistors since 2022, three generations deep now, while TSMC is only just bringing its own first GAA node to market this year. Being first to the structure already happened, back at 3nm, and it earned Samsung almost nothing. So the SF2 pitch leans on the one asset that lead actually produced: maturity. Three years of making every mistake this transistor has to offer, three years of learning to yield nanosheets, handed to a customer who gets to skip the climb. It is the strongest case Samsung has built in years, because the engineering lead finally lines up with what a customer actually pays for, a process that works on day one.
What Samsung carries into 2nm is a trust deficit, and it has nothing to do with the transistor. A decade of nodes where TSMC’s silicon, right up through 3nm, did exactly what it promised while Samsung’s kept surprising customers in expensive ways, that is the actual ledger. A foundry’s entire product is the promise that its process will not be the reason your chip fails, and Samsung has broken that promise loudly enough, and recently enough, that no quantity of third-generation-GAA confidence clears it on its own. SF2 can be the best silicon Samsung has ever shipped and still lose, because the hole 4nm dug was never in the cleanroom. It was in whether a customer will risk a billion-dollar tape-out on a Samsung promise again.
Qualcomm got burned at Samsung once and is comfortable at TSMC now. Nvidia, riding the AI wave, needs the most reliable process it can buy and has zero reason to gamble a flagship accelerator on a second source. Those are the candidates, and the list of companies that could validate SF2 the way Apple validates TSMC basically stops there. The real money now sits in AI accelerators, the giant chips that train and run the models, and those buyers are the most risk-averse customers in the entire industry, because each design costs a fortune and is sold out years before it tapes out. One of them on SF2 would do more for Samsung than any benchmark slide ever could. All of them staying with TSMC, with Intel clawing at whatever is left, is the quiet way 2nm becomes one more Samsung first that never paid.
Intel is now pushing its own gate-all-around process with backside power delivery, planting itself as a third leading-edge foundry and going after the exact same short list of customers. For Samsung that cuts both ways, and the bad way is bigger. Yes, a credible Intel muddies the TSMC monopoly everyone has been afraid of. It also means Samsung is no longer the default runner-up; it is now scrapping for second place against a desperate, government-backed competitor that needs those wins just as badly. TSMC sits out front, and behind it two wounded companies are fighting over the same scraps. The automatic-alternative slot Samsung used to own is gone.
Whether SF2 is the actual turn or just the next chapter of the same story is genuinely unknown right now, and anyone who tells you they are sure is guessing. Perso, I want it to work. A real three-way fight at the leading edge beats a TSMC monopoly for everyone who buys or uses a chip, and the maturity argument is the most honest thing Samsung has said about itself in a decade. But wanting a thing and betting on it are not the same, and Samsung’s history is hard to argue with. It got to FinFET early, beat everyone to EUV, shipped gate-all-around first, and finished behind on the only metric a foundry actually sells, which is whether you can trust it. SF2 is Samsung asking the industry to trust that the experience finally outweighs the track record. That call belongs to a dozen risk-averse customers, and so far, every time it has mattered, they have made it against Samsung.