Login to Continue Learning
NVIDIA is reportedly set to be TSMC’s first customer for the A16 node, marking a significant shift in the firm’s approach. Traditionally, NVIDIA hasn’t been among the earliest adopters of TSMC’s cutting-edge processes, with companies like Apple, MediaTek, and Qualcomm often leading this trend.
NVIDIA’s Pivot Towards TSMC’s Latest Processes Comes After Two Decades, Amid Huge Competition in the AI Space
NVIDIA plans to improve the performance of its future AI architectures by leveraging TSMC’s latest processes. Historically, NVIDIA has focused on architectural advancements rather than being an early adopter of TSMC’s most recent nodes. According to a report by Ctee, NVIDIA will be the first to use the A16 process, which is expected to offer significant performance improvements and align with Moore’s Law.
Since 110nm, NVIDIA hasn’t been the first to rely on TSMC’s cutting-edge nodes. Instead, the company has focused on architectural advancements for scaling performance. The A16 process will likely be a major release from TSMC, driven by the introduction of technologies such as Gate-All-Around FETs (GAAFET) and Super Power Rail (SPR) backside power delivery. Among these advancements, SFR is particularly significant.
NVIDIA’s transition to the A16 node could occur in late 2027 or early 2028, considering that TSMC plans high-volume manufacturing (HVM) by late 2026. Based on NVIDIA’s roadmap, this technology might debut with either the Rubin Ultra or Feynman GPUs, but the latter is more likely due to additional advancements in Feynman.
This move is good news for TSMC, as it will enable mainstream AI hardware manufacturers to use the company’s high-end nodes. This will ultimately contribute to revenue from the process generation.
