Home > Business > Industry

Samsung becomes Nvidia foundry partner as AI chip opportunities grow

Published: 14 Oct. 2025, 19:09

Nvidia CEO Jensen Huang speaks about new products as he delivers the keynote address at the GTC AI Conference in San Jose, California, on March 18. [AFP/YONHAP]

Nvidia has unveiled new technology that connects even third-party chips into its massive AI data centers — and in doing so, added Samsung Electronics’ foundry business to its list of partners.

As Google, Meta, Amazon and OpenAI push forward with custom-built AI semiconductors, Nvidia has responded by announcing it will enable those chips to work seamlessly with its own graphics processing units (GPUs). The inclusion of Samsung in this ecosystem gives the company a new opportunity to secure high-profile foundry clients.

AI data centers are getting bigger

On Monday, Nvidia revealed next-generation hardware and platforms for large-scale AI data centers at the Open Compute Project (OCP) Global Summit in San Jose, California.

Nvidia currently offers the NVL72 server, which connects 72 Blackwell GPUs. Its next-generation Rubin architecture will power the NVL144, which connects 144 GPUs.

The Rubin NVL144 system uses 100 percent liquid cooling and can also be paired with the Rubin CPX — a specialized GPU that uses GDDR7 memory instead of high bandwidth memory (HBM) to enable more cost-effective large-scale AI inference.

By 2027, Nvidia aims to launch a GPU-based AI supercomputer called Kyber, which will pack 576 GPUs into a single system that operates like one unified machine.

A notice on Nvidia's homepage shows that the company is adding Inteland Samsung Electronics to its NVLink Fusion partnerships [SCREEN CAPTURE]

Nvidia to open proprietary tech to third parties

At the heart of this scale-up is connection speed. Nvidia owns proprietary interconnect technologies like NVLink and Spectrum-X, which allow ultrafast communication between chips and servers.

As AI models grow in size and demand more infrastructure, Nvidia’s networking edge has helped it retain over 90 percent market share, outpacing rivals like AMD.

At the same time, more companies are building their own AI chips to reduce reliance on Nvidia. On Monday, OpenAI announced it was working with Broadcom to develop a 10-gigawatt-scale custom AI chip. OpenAI will handle the chip design, and Broadcom will lead development and manufacturing. The foundry has not been disclosed, but Reuters has previously reported that OpenAI is in talks with TSMC.

To keep these clients from leaving entirely, Nvidia introduced NVLink Fusion in May — a new version of NVLink that selectively opens up its interconnect tech to outside companies. This allows third-party chips to be integrated into Nvidia GPU-based data centers. In effect, even chips built to be independent of Nvidia would still operate within the “Nvidia ecosystem.”

Samsung Electronics Chairman Lee Jae-yong and Nvidia CEO Jensen Huang embrace during a Korea-U.S. business roundtable at the InterContinental the Willard Washington in Washington on Aug. 25. [JOINT PRESS CORPS]

What the Samsung partnership means

With AI companies increasingly turning to tailor-made chips and TSMC’s foundry capacity already maxed out, Samsung now has a chance to win new clients.

At the OCP summit, Nvidia officially announced that Intel and Samsung Foundry had joined the NVLink Fusion ecosystem. Nvidia will work with Intel to develop X86 CPUs using NVLink Fusion, according to the tech giant.

“Samsung Foundry has partnered with Nvidia to meet growing demand for custom CPUs and custom XPUs, offering design-to-manufacturing experience for custom silicon,” said Nvidia in a statement.

Existing NVLink Fusion partners include CPU makers like Fujitsu and Qualcomm, as well as custom chip designers like Broadcom and Marvell. Samsung Foundry’s inclusion means that if OpenAI or other firms build custom chips with Samsung, those chips will be able to interface with Nvidia’s platform via NVLink Fusion.

At the summit, Nvidia also announced that Meta and Oracle would adopt Spectrum-X for their AI data centers.

Nvidia CEO Jensen Haung said at the OCP summit that AI models now operate at the scale of trillions of parameters, turning data centers into gigascale AI factories. He added that Spectrum-X is the “nervous system” that connects millions of GPUs inside those factories.

This article was originally written in Korean and translated by a bilingual reporter with the help of generative AI tools. It was then edited by a native English-speaking editor. All AI-assisted translations are reviewed and refined by our newsroom.
BY SHIM SEO-HYUN [[email protected]]