As manufacturing demand from big tech corporations concentrates on TSMC's advanced nodes, bottlenecks are worsening, and even the top client Nvidia is reportedly set to adjust output. The plan is to cut production of the top-tier model of the next-generation artificial intelligence (AI) accelerator "Rubin Ultra," slated for release next year, while increasing output of the previous-generation Blackwell platform.
According to the industry on the 2nd, Nvidia is said to be scaling back production of Rubin Ultra, which is scheduled for mass production on TSMC's 3-nanometer (nm; one-billionth of a meter) process, compared with the original plan. Instead, it will reportedly increase production of the Blackwell platform, made on 4 nm. TSMC is expanding capacity for its cutting-edge 3 nm process now in mass production, but demand is flocking from clients such as Apple, making on-time delivery difficult.
Bottlenecks have deepened as Samsung Electronics' foundry division, a TSMC competitor, has struggled to secure clients at nodes 3 nm and below due to weak Production yield. On top of that, the AI industry's rapid growth has caused an explosion in AI chip manufacturing demand. Unlike the past, when smartphone application processor (AP) clients such as Apple and Qualcomm accounted for a large share of volume, the increase now comes from big tech corporations like Google, Amazon, and Meta seeking to manufacture AI chips.
Amid this, with TSMC's capacity constrained and on-time delivery looking unlikely, Nvidia also appears to be moving to adjust output. Starting this year, Nvidia will mass-produce the AI chip Rubin, which adopts TSMC's 3 nm process for the first time. The previous-generation Blackwell platform is mass-produced on the 4 nm process.
Nvidia is TSMC's largest client, and the fact that even Nvidia adjusted output indicates how severe TSMC's capacity constraints are. Nvidia is seen as adjusting the mass-production schedule and output of the top-end Rubin version, while responding by expanding volumes of the Blackwell platform, which is mass-produced on the relatively less-loaded 4 nm line.
Some say the bottleneck from limited capacity is becoming a reality—TSMC has said its processes are "fully booked" through 2028—creating an opening for Samsung Electronics' foundry division. Although weak Production yield at Samsung Electronics' foundry division had tightened the tilt toward TSMC, the Production yield of advanced nodes at 3 nm and below is now said to have stabilized.
In fact, Tesla has formalized 2 nm collaboration with Samsung Electronics' foundry division for AI chip production, and global big tech corporations such as Qualcomm and AMD are reportedly exploring 2 nm cooperation with Samsung Electronics' foundry division. Groq, an AI Semiconductor startup that Nvidia acquired indirectly, also asked Samsung Electronics' foundry division to ramp up production.
A semiconductor industry official said, "Concerns were raised two to three years ago that bottlenecks could intensify as TSMC effectively monopolized advanced nodes, and that is becoming reality amid the AI industry's steep growth," adding, "At this point, as Samsung Electronics' advanced-node Production yield stabilizes, it is becoming an opportunity to win volume from big tech corporations."