Huang Jensen, Nvidia CEO, delivers the keynote at the annual developer conference GTC 2026 at the SAP Center in San Jose, California, on the 16th (local time). /Courtesy of AP Yonhap

Nvidia CEO Jensen Huang emphasized cooperation between the two companies by separately mentioning and thanking Samsung Electronics during his keynote at the annual developer conference, GTC2026. In particular, he put forward Samsung Electronics as a key partner for manufacturing Nvidia's language processing unit (LPU) chip, "Groq 3."

Huang, during his keynote on the 16th (local time) at the SAP Center in San Jose, California, introduced the inference-only chip and said, "Samsung is manufacturing the Groq 3 LPU chip for us," adding, "We are ramping up production as fast as we can right now. We are truly grateful to Samsung."

He explained that the chip will be mounted in Nvidia's next-generation AI chip system "Vera Rubin," saying, "Shipments will likely begin in the second half of this year, probably around the third quarter." The Groq 3 LPU is a chip that splits roles with Nvidia's "Rubin" graphics processing unit (GPU) to improve inference performance and efficiency, and Huang's remarks confirmed that Samsung Electronics' foundry (contract chip manufacturing) division is producing it.

Samsung Electronics also actively promoted its cooperation with Nvidia in memory by unveiling to the public for the first time the physical chip of its next-generation HBM, "HBM4E," and the "core die" wafer for stacking at its exhibition booth at the GTC venue.

Aiming for sample shipments in the second half of this year, HBM4E is expected to support a per-pin transfer rate of 16 Gbps (gigabits per second) and bandwidth of 4.0 TB/s (terabytes per second). This surpasses the 13 Gbps transfer rate and 3.3 TB/s bandwidth of the latest 6th-generation HBM4, which began mass-production shipments on the in the previous month.

Samsung Electronics plans to accelerate HBM4E development based on the 1c (10-nanometer-class 6th-generation) DRAM process technology competitiveness accumulated through HBM4 mass production and the design capabilities of Samsung Foundry's 4-nanometer base die (the core component mounted at the very bottom of HBM).

Samsung Electronics' decision to unveil HBM4E immediately after starting mass-production shipments of HBM4 is seen as an attempt to highlight its lead over other competitors such as SK hynix and Micron. Among major suppliers, Samsung Electronics is the only memory company to put next-generation HBM4E front and center at this GTC.

In particular, Samsung Electronics used the "HBM4 Hero Wall" at this exhibition to assert that such competitiveness is a strength unique to an integrated device manufacturer (IDM). Samsung Electronics also unveiled, via video, its hybrid copper bonding (HCB) packaging technology, which improves thermal resistance by 20% compared with thermo-compression bonding (TCB)—a method that connects chips stacked with heat and pressure—and supports ultra-high stacking of 16 or more layers.

※ This article has been translated by AI. Share your feedback here.