OpenAI's self-developed artificial intelligence (AI) inference chip, Jalapeño, will incorporate Samsung Electronics' sixth-generation high bandwidth memory (HBM4). While the possibility of Samsung supplying HBM4 had been raised recently, the actual supply details had not been disclosed until now. The confirmation that Samsung's HBM4 will be used in OpenAI's AI chip is expected to further diversify the AI semiconductor supply chain.
According to industry sources on August 27, the HBM4 to be used in Jalapeño has been confirmed as a Samsung product. Each chip will feature 216GB of HBM4, with a reported memory bandwidth of 15.4TB/s.
Jalapeño is an AI accelerator developed by OpenAI in collaboration with Broadcom. OpenAI plans to deploy Jalapeño in its infrastructure starting later this year. The company is also reportedly pursuing the development of subsequent chip generations.
This supply agreement marks a shift for Samsung, expanding its HBM customer base from traditional accelerator companies like Nvidia to big tech firms developing their own AI chips. Earlier this year, Samsung outlined a strategy to secure HBM customers not only from global GPU manufacturers but also from hyperscalers creating their own chips. The confirmation of HBM4 supply to OpenAI indicates that this diversification strategy is taking shape.
OpenAI's development of its own chips is seen as a new opportunity for the AI semiconductor market. Analysts suggest that if big tech companies design accelerators optimized for their services rather than relying solely on Nvidia GPUs for all AI computations, the AI chip market could evolve into a structure where general-purpose GPUs coexist with custom application-specific integrated circuits (ASICs). An increase in proprietary AI chips could also lead to a rise in demand for HBM.
For Samsung, this development is viewed positively. An increase in HBM customers could reduce reliance on specific clients. If competition among customers emerges, it could also enhance Samsung's negotiating power regarding prices. As the AI accelerator market expands, the addition of big tech's proprietary chips as HBM customers is expected to bolster Samsung's HBM shipment volumes.
In addition to HBM4, Samsung is also pursuing the development of customized HBM tailored to individual customer requirements. Industry experts view the supply to OpenAI as a potential example of future collaborations between big tech and memory manufacturers that could extend beyond simple product supply to include AI chip design stages.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.