[News] NVIDIA GTC 2026 in Focus: Feynman Reportedly on TSMC A16, Samsung & SK hynix to Showcase HBM4

Ahead of NVIDIA’s earnings release, the industry’s gaze is already fixed on its mid-March’s GTC conference. CEO Jensen Huang’s tease of a “world-surprising” new chip has set the stage for a showcase spanning both foundry and memory. According to Chosun Biz, the chip, likely the upcoming Feynman processor, is reportedly built on TSMC’s A16 node, while key HBM4 suppliers for the Vera Rubin platform — Samsung Electronics and SK hynix — are also set to take center stage at the event.

Feynman Foundry Partner in Focus

As reported by Commercial Times, with NVIDIA having confirmed at CES that the Vera Rubin platform has entered mass production, industry watchers are eyeing two possibilities for GTC: the company could unveil Rubin-based extensions, including new GPUs or related compute chips, or it might drop an early glimpse of its next-generation Feynman architecture.

The Economic Daily News notes that, in the first scenario, NVIDIA is likely to elaborate on Rubin R100, expected to use a 3nm process paired with HBM4. In the second scenario, Team Green could showcase the Feynman architecture or a technical prototype, which leverages silicon photonics for optical interconnects between rack-scale compute units, the report adds.

Feynman, currently under development and expected to launch in 2028, is reportedly built on TSMC’s A16 node, marking NVIDIA’s first mass-produced chip using a 1nm-class process, according to Chosun Biz. However, the report also brings up speculation that NVIDIA may outsource part of Feynman GPU production to Intel, with talks covering both foundry manufacturing and advanced packaging.

According to a previous Wccftech report, big tech firms such as NVIDIA are considering outsourcing less complex, lower-risk chips to Intel—such as Feynman’s I/O die on 14A or 18A, along with EMIB advanced packaging—to protect volume ramp-up in case Intel falls short on yield or capacity, while keeping GPU core production anchored at TSMC.

HBM4 Suppliers Set to Steal the Spotlight

Another highlight at GTC will be how NVIDIA’s key memory partners showcase their latest breakthroughs. Chosun Biz reports that Samsung Electronics will spotlight the performance of its HBM4, officially shipped on Feb. 12 and reportedly mass-produced for NVIDIA’s Rubin GPUs. The company is also expected to share insights on its next-generation HBM4E and how it will interface with future NVIDIA GPU platforms, the report adds.

SK hynix is likewise scheduled to introduce its HBM4 technology at GTC, presenting under the theme of how HBM4 improves the efficiency of large language models (LLMs), as per Chosun Biz. The company has confirmed it is already mass-producing HBM4 for NVIDIA’s Rubin GPUs and is expected to supply roughly two-thirds of NVIDIA’s total HBM4 demand this year, the report notes.

TrendForce observes that Samsung currently leads in terms of validation progress and is expected to begin phased mass production following completion in 2Q26. SK hynix continues to advance steadily and is likely to retain a competitive edge in bit allocation, supported by its longstanding HBM partnership with NVIDIA. Micron, while progressing at a relatively slower pace, is also expected to complete validation by 2Q26, according to TrendForce.

Read more

• [News] NVIDIA May Relax HBM4 Specs as Samsung and SK hynix Reportedly Face Capacity, Yield Limits
• [News] NVIDIA May Be TSMC’s Only A16 Customer So Far While Apple Reportedly Yet to Start Talks

(Photo credit: NVIDIA)