Samsung Foundry is trying to dig itself out of the hole it fell into back in 2022 when a low yield rate of 35% led Qualcomm to push all of its chips in on TSMC. The latter had a yield of 70% at the time. Yield remains an issue today for Samsung Foundry and might even determine whether the Galaxy S26 line, due out early next year, includes models powered by Sammy's own Exynos 2600 application processor (AP) or a more expensive Snapdragon AP.
Last year, Samsung Foundry's inability to make enough Exynos 2500 APs due to Samsung Foundry's low yield cost Sammy a ton of money. The original plan was for Samsung to power the base Galaxy S25 and the Galaxy S25+ models in most markets with the homegrown Exynos 2500 AP. Unable to build as many Exynos 2500 SoCs as it needed, Samsung ended up using Qualcomm's Snapdragon 8 Elite for Galaxy APs on all models which cost Samsung an extra $400 million that it hadn't expected to need.
Samsung faces a similar issue with next year's Galaxy S26 and Galaxy S26+ models. The difference this time is whether the company's foundry can manufacture enough Exynos 2600 chipsets. A foundry's yield is the percentage of chips manufactured from a silicon wafer that are defect-free and have passed Quality Control. Most foundries aim for a 70% yield before they start to mass produce a certain SoC. Low yields can lead to higher manufacturing costs and result in a customer paying higher prices for their chips.
Samsung Foundry is trying to make a comeback. As we noted, it hopes to improve its 2nm yield fast enough to produce enough Exynos 2600 APs to power the Galaxy S26 and Galaxy S26+ in most markets except for the U.S., China, and Canada. That could make those phones the first to employ 2nm chipsets.
A report out of South Korea today suggests that Samsung Foundry is hoping to get a contract from Nvidia to build its next-gen GPU (Graphics Processing Unit) using its 2nm process node. Because GPUs use parallel processing allowing them to perform many tasks simultaneously, they are more suitable for use as AI accelerators than CPUs (Central Processing Units) are. CPU's use sequential processing which make them less useful for AI. AI accelerators significantly speed up the processing of Artificial Intelligence and Machine Learning workloads.
The latest yield rate for Samsung Foundry's 2nm process node is 40% and by the end of the year, it is expected to hit the level that will allow the foundry to become a legitimate alternative to TSMC. Getting Nvidia's GPU business could help Samsung Foundry's reputation (which needs a shot in the arm) and its market share. At 7.7% during Q1 2025, Samsung Foundry is well behind TSMC's global market share of approximately 67.6%.
Samsung Foundry might already be on the rise as it built the Nvidia designed Tegra T239 SoC that powers the popular Nintendo Switch 2 game console. The foundry used its mature 8nm Deep Ultraviolet (DUV) process node to manufacture the component taking the Tegra business away from TSMC.