As large language models and rack-scale supercomputers grow exponentially, the good old method of pairing off-the-shelf processors with separate memory banks is no longer an option. AI workloads today need tightly integrated solutions where compute and memory are co-engineered from the ground up. However, with the exponential rise in AI usage across the globe, there"s also a huge supply bottleneck as well as bandwidth limitations that have left AI companies yearning for more suitable hardware for such inference-heavy AI jobs. This is why AMD and Samsung are coming together and have signed a Memorandum of Understanding (MOU) to help make the next generation of hardware needed for supporting AI infrastructure.
The MOU was signed at Samsung’s state-of-the-art chip manufacturing complex in Pyeongtaek, Korea, by AMD Chair and CEO Dr. Lisa Su, alongside Samsung Vice Chairman and CEO Young Hyun Jun. According to the agreement, Samsung would be the primary supplier of next-generation HBM4 memory for AMD’s forthcoming Instinct MI455X AI accelerators.
The partnership also includes the CPU market, where Samsung will provide highly optimized DDR5 memory for AMD’s 6th Gen EPYC processors, internally codenamed "Venice." Both of these components will serve as the backbone of AMD’s ambitious Helios rack-scale architecture.
The partnership comes at a time when AI has already triggered a massive pivot toward high-bandwidth memory, driving up costs across the broader consumer storage and DRAM markets. For Samsung, this is a great opportunity to double down on its most profitable sector and become a gatekeeper in the AI sector.
For AMD, this partnership ensures that its hardware will have the memory headroom required to power increasingly complex software. Team Red has been ******* big on AI, rolling out a robust suite of local inference tools and AI-focused hardware earlier this year to rival cloud-based models. A steady supply of HBM4 will ensure that their software stack keeps evolving as enterprise clients demand larger parameter models.
Not only this, but Samsung and AMD are also actively discussing opportunities for a deeper foundry partnership, which would see Samsung fabricating next-generation AMD products.
While it is still unclear what the scope of this foundry would include, the two companies have already collaborated on everything from mobile smartphone processors utilizing the RDNA 2-based Xclipse GPU to their present-day arrangement, where Samsung provides HBM3E memory for the Instinct MI350X and MI355X accelerators.