Samsung Foundry's woes might finally come to an end in the coming months. An earlier report stated its 2 nm yields had improved to 40-50%, but it came at the price of performance. Nevertheless, this hasn't deterred major players from giving the chipmaker a chance. Qualcomm might reignite its partnership with Samsung Foundry for the Snapdragon 8 Elite 2 for Galaxy, and if a recent report from Chosun is accurate, so could Nvidia.
It states an upcoming Nvidia GPU could be manufactured on Samsung Foundry's 2 nm node. Unfortunately, it doesn't say which one (SF2/SF2P), and at this point, it is far too early to speculate. Nvidia's most profitable AI/datacentre SKUs are unlikely candidates because of the company's robust partnership with TSMC. Additionally, an alleged TSMC roadmap from earlier showed some of its N2 (2 nm) clients, and that included Nvidia.
It leaves us with the next generation of RTX 60 series gaming GPUs, powered by the Rubin architecture. Nvidia has already worked with Samsung Foundry's 8N node to make Ampere (RTX 30 series) chips. The transition from a Fin FET-based TSMC 4NP to a GAA FET-based SF2 might result in some performance improvements, but they're unlikely to be as impressive as the jump between Ampere and Ada Lovelace.
That said, there's a solid chance Nvidia could use Samsung's nodes for its future laptop GPUs or Windows-on-Arm offerings. Laptop chips are usually much smaller in size and hence easier to manufacture. For example, the Snapdragon X Elite has an approximate die area of 170 mm2, while Nvidia's flagship GB202 GPU measures a whopping 750 mm2. Large chip like the latter are typically made on mature nodes, and SF2/SF2P have not been around long enough to qualify as one.
Nvidia and Qualcomm will essentially determine the future of Samsung's Foundry business in the coming months. At this point, the chipmaker doesn't even need to outdo TSMC—neck-and-neck performance will suffice. And if things do work out for the better, more contracts may be on the cards because Samsung, despite its woes, has a roadmap that keeps up with its competitors. Its 2027-bound SF2Z node has backside power delivery, and its 1.4 nm SF1.4 is scheduled around when TSMC plans to unveil its own A14 node.