[News] NVIDIA Reaffirms TSMC as Preferred U.S. Advanced Packaging Partner, Reportedly Sidelines Intel

During the international press conference for Computex 2025,  NVIDIA CEO Jensen Huang, when asked whether NVIDIA might use Samsung or Intel for advanced packaging in the U.S., said the technology is highly advanced and that NVIDIA currently has no alternative—reaffirming TSMC as its sole partner, according to Economic Daily News.

As noted by the report, Huang emphasized NVIDIA’s focus on advanced packaging, citing the Grace Hopper Superchip as an example. Its size exceeds optical lithography limits by nearly twofold, prompting NVIDIA to use TSMC’s CoWoS technology to integrate and package the components, as the report notes.

Huang further explained that NVIDIA integrates multiple chiplets into one or more packaging modules, which can include memory, storage, and other computing components—such as silicon photonics, atomic storage, or various heterogeneous elements—within a single package, the report indicates.

MediaTek Joins NVIDIA in Advancing Desktop AI Supercomputers

At the same time, the report states that limited HBM output is affecting AI server development. In response, NVIDIA has partnered with MediaTek to develop AI supercomputers using the more mature and widely available LPDDR memory.

Huang explained that current AI systems are constrained by limited accessible memory, prompting NVIDIA to develop the Grace CPU—the first AI chip designed to work with large volumes of LPDDR, typically used in mobile devices. NVIDIA’s RTX Spark (previously known as Project Digits) is also the first AI supercomputer built with LPDDR memory, as the report highlights.

NVIDIA’s DGX Spark is a compact, high-performance desktop AI supercomputer, which is already in full production and expected to launch soon, as noted by Anue.

Read more

• [News] Jensen Huang: NVIDIA’s China Market Share Plummets from 95% to 50% amid U.S. Export Curbs
• [News] NVIDIA Unveils NVLink Fusion, Amplifying ASIC Push with MediaTek & Marvell Onboard

(Photo credit: NVIDIA)