After a rough stretch that saw Samsung lose ground to TSMC in foundry and to SK Hynix in memory, the company’s semiconductor business is showing its most convincing signs of recovery in years. Samsung Foundry’s utilization has climbed to somewhere between 70% and 80% and is on track to hit 100% in the second half of 2026, according to industry tracker TrendForce, driven by surging orders for 2-nanometer chips, AI accelerators and HBM base dies.
Customers lining up for 2nm
The most telling detail in Samsung’s turnaround is who is reportedly showing up for its 2nm process. According to reporting picked up by SamMobile and Sammyfans, Qualcomm, AMD, Google and Tesla are all said to be in discussions for future chip production on Samsung’s advanced node — a roster that, if it materializes into firm orders, would mark a meaningful diversification away from Samsung’s historical reliance on its own internal Exynos and memory businesses for foundry volume. Executives and industry sources have been quoted describing the sentiment among Samsung’s chip clients simply as: Samsung is back.
zHBM: a bid to leapfrog the memory competition
On the memory side, Samsung used the FMS 2026 conference to unveil zHBM, a 3D memory architecture that stacks memory directly atop AI chips rather than placing it alongside them on a substrate. According to Samsung’s own announcement and reporting from Seoul Economic Daily, zHBM delivers performance up to eight times that of HBM5, the next-generation memory standard, while improving performance-per-watt by up to three times. If those figures hold up in real-world AI accelerator designs, zHBM would represent a genuine architectural leap rather than an incremental speed bump, since 3D-stacking memory directly onto compute dies fundamentally shortens the data path between memory and processor.
The $200 billion partnership with Broadcom
Samsung’s momentum got a further boost from an expanded strategic collaboration with Broadcom, announced via Samsung’s global newsroom, that the two companies estimate could be worth more than $200 billion across memory and foundry over the next five years through 2030. The memory side of the deal centers on HBM supply, while the foundry component focuses specifically on Samsung’s 2-nanometer-and-below process technology for Broadcom’s custom AI chip products — the same custom silicon business that Broadcom CEO Hock Tan has said will exceed $100 billion in annual revenue by 2027.
Reasons for skepticism
Not every analyst is ready to call this a full turnaround. Samsung has announced ambitious foundry roadmaps before that underdelivered against TSMC’s execution, and 2nm yield data at commercial volumes remains largely unproven outside company statements. Some industry observers note that being in discussions with Qualcomm, AMD, Google and Tesla is a far cry from signed, binding wafer commitments, and that TSMC retains a multi-year lead in both leading-edge yield and customer trust built over a decade of reliable delivery. Skeptics also point out that Samsung’s memory business still trails SK Hynix in HBM market share, with SK Hynix holding roughly 62% of the HBM market as of mid-2026.
What’s next
The real test for Samsung comes in the back half of 2026: whether foundry utilization actually reaches the 100% target, whether any of the reported 2nm prospects convert into announced customer wins the way Fortinet’s deal validated Intel’s foundry ambitions, and whether zHBM moves from conference demo to qualified production memory that AI chip designers actually adopt. If Samsung can convert reported interest into signed contracts across both its foundry and memory businesses simultaneously, it would mark the first time in several years that the company credibly competes at the frontier on two fronts at once, rather than playing catch-up in one while defending share in the other.
Executives at Samsung’s device solutions division have said internally that regaining customer trust after several years of yield and delivery setbacks is as important as the underlying technology itself, since large chip customers typically qualify a second or third supplier only after extended reliability track records, and Samsung’s own history of yield stumbles on prior advanced nodes means prospective clients are watching early 2nm production data unusually closely before committing meaningful order volume. The next two quarters of earnings calls, where Samsung will need to show utilization figures and named customer commitments rather than reported interest, will be the clearest test yet of whether the recovery narrative holds.