SK hynix ships early HBM4 samples to Nvidia

Home > Business > Tech

SK hynix ships early HBM4 samples to Nvidia

Published: 19 Mar. 2025, 10:31 Updated: 19 Mar. 2025, 18:45

JIN EUN-SOO
[email protected]

Audio report: written by reporters, read by AI

SK hynix's HBM4 chip [SK HYNIX]

SK hynix is shipping samples of its latest high bandwidth memory (HBM) chips, solidifying its standing in the competitive AI memory sector.

The Korean chipmaker announced Wednesday that it had started supplying 12-high HBM4 samples to clients including Nvidia.

“SK hynix is starting the verification process of 12-high HBM4 products with clients, as we started shipping out samples earlier than expected on the back of our technological edge and extensive experience in the HBM industry,” the company said in a release.

“Mass production preparations are expected to be completed later this year.”

As the sixth generation of its kind, the 12-high HBM4 can process 2TB of data per second, the equivalent of more than 400 FHD movies. This represents a 60 percent speed increase over its predecessor, the HBM3E, according to SK hynix.

“We have enhanced our position as a front-runner in the AI ecosystem following years of consistent efforts to overcome technological challenges in accordance with customer demands,” said Kim Ju-seon, president and head of AI infrastructure at SK hynix.

“We are now ready to smoothly proceed with the performance certification and preparatory works for mass production, taking advantage of the experience we have built as the industry’s largest HBM provider.”

SK hynix also unveiled a mock-up of its 12-high HBM4 for the first time at Nvidia's GTC event in San Jose, California, one of the biggest events for AI and semiconductor players.

It exhibited an array of AI memory solutions, such as those for AI data centers, on-device applications and vehicles at its booth titled “Memory, Powering AI and Tomorrow.”

At the event, Nvidia revealed a detailed road map for its AI processors, including Rubin, the successor to the Blackwell series, slated for launch in the latter half of 2026. Rubin will be equipped with HBM4 memory.

Samsung Electronics, a runner-up in the AI memory chip realm, is also going all-in to prove its competitiveness to the global tech industry. It is holding a tech session for its newest seventh-generation graphic double data rate (GDDR) memory, or GDDR7, at the event in San Jose, according to JoongAng Ilbo.

The Korean chip giant, the main supplier of GDDR7 for Nvidia's GPUs, including its GeForce RTX 50 series, aims to maintain a presence at the conference as a legitimate supplier to the U.S. chip giant, something it wasn't able to do in the realm of HBM.

Samsung Electronics will also showcase its latest optical proximity correction technology, which adjusts the pattern on the semiconductor paired with extreme ultraviolet lithography machines.