NVIDIA and Samsung in Talks for Advanced AI Memory Chip Development

NVIDIA and Samsung are reportedly engaged in confidential negotiations to commercialize a new type of advanced AI memory chip known as SOCAMM (system on chip advanced memory module). These discussions come as NVIDIA seeks to enhance its AI accelerators, which currently rely heavily on high-bandwidth memory primarily supplied by SK Hynix.

The SOCAMM chip is seen as an advancement over existing high-bandwidth memory technologies. It is designed to improve the performance of artificial intelligence supercomputers by offering a higher cost-performance ratio and a modular design that reduces data bottlenecks in AI computing.

Also, SOCAMM’s detachable architecture allows data center operators to easily swap and upgrade memory modules, allowing ongoing performance enhancements.

Both companies are already exchanging prototypes of the SOCAMM chip and conducting performance tests, with mass production expected to commence before the end of 2025. This collaboration could provide Samsung with a big opportunity to regain market share in the high-bandwidth memory segment, an area where it has faced challenges against SK Hynix.