Samsung Electronics will be the first to exclusively supply its next-generation High Bandwidth Memory (HBM4) to OpenAI, the world’s largest artificial intelligence (AI) company based in the U.S. OpenAI plans to integrate Samsung’s HBM4 into its first-generation AI semiconductor, Titan. With Samsung now supplying HBM4 to OpenAI in addition to Nvidia, it is assessed that the company is solidifying its leadership in the advanced AI chip market.

According to industry sources on March 19, Samsung Electronics has agreed to exclusively supply OpenAI with up to 800 million gigabits (Gb) of HBM4 (12-stack product) in the second half of this year. This volume accounts for 7% of Samsung’s total planned HBM production for the year (over 11 billion Gb). Based on its flagship HBM4 product (5.5 billion Gb), this means approximately 15% of the volume is allocated to OpenAI. This is reported to be the third-largest supply volume, following deals with Nvidia and AMD.

The HBM4 supplied by Samsung will be placed directly next to Titan, OpenAI’s first AI semiconductor set to be released this year. Titan is a specialized AI chip developed in collaboration with the world-renowned semiconductor design company Broadcom. It has been reported that TSMC will begin production in the third quarter of this year for a release around the end of the year.

The analysis suggests that Samsung’s exclusive HBM4 supply deal is significant because it was secured with OpenAI, the AI company that has garnered the most attention from the global information technology (IT) industry. OpenAI is the company that sparked the generative AI boom with the release of ChatGPT in 2022. It plays a central role in the Stargate Project, a $500 billion AI infrastructure initiative planned by the United States.

OpenAI provides its generative AI services by installing hundreds of thousands of artificial intelligence (AI) semiconductors in its data centers. Until now, it has primarily relied on Nvidia’s general-purpose AI semiconductors. Recently, OpenAI determined that it needs its own custom semiconductor, rather than Nvidia’s general-purpose AI chips, to more efficiently implement its own inference models. This comes as inference performance, which involves making judgments based on learned data, has become the dominant trend in AI.

An industry official commented, “OpenAI is pouring efforts into research and development (R&D) to succeed in the mass production of its proprietary chip, Titan. Samsung Electronics successfully secured this supply deal by meeting the stringent conditions for 6th-generation High Bandwidth Memory (HBM4) presented by OpenAI.”

There are also observations that since OpenAI chose Samsung as its first HBM supplier, there is a high probability that Samsung’s HBM will also be integrated into future generations of the Titan chip, such as the 2nd and 3rd generations.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved