Samsung's foundry unit is profitable again, but its most advanced chips still aren't good enough, fast enough, or cheap enough to close the gap with TSMC.
Samsung began mass-producing its first 2nm chips in 2025 and moved equipment into its long-delayed Taylor, Texas fab in April, where a third-generation 2nm variant is due for trial production by the end of 2026 and full output in 2027. The plant's anchor client is Tesla, whose AI6 chip is covered by a $16.5 billion contract, though that chip has reportedly slipped about six months because of a delayed engineering run on Samsung's line. Meanwhile 2nm yields are still sitting near 55%, below the level Samsung needs to run the node profitably, and the company has responded by cutting its 2nm wafer price to around $20,000, roughly a third cheaper than TSMC. Samsung's own Exynos 2600 anchors the node commercially but reportedly supplies only a quarter to a third of Galaxy S26 units because low yields cap output.
The yield problem is the real story here, not the fabs or the customer list. Samsung can absorb these losses only because its foundry sits inside a division that also holds a memory business currently posting record profits from HBM4; a pure-play foundry wouldn't have that cushion. That structural advantage is also why Samsung just pushed its 1.4nm node from a 2027 target out to 2029, buying three more years to fix the current process before moving down a node.
Undercutting TSMC on price while trailing it on yield isn't a sustainable position. It's a subsidy, and it's memory, not logic, paying for it.