Table of Contents
AMD wants to secure its memory supply in the face of massive shortages
In sum – what we know:
- A multifaceted alliance – Samsung will provide HBM4 for AMD’s MI455X accelerators and DDR5 for EPYC processors, while exploring a future as a contract manufacturer for AMD chips.
- Securing the supply chain – The deal addresses critical HBM shortages and supports AMD’s massive delivery commitments to partners like Meta Platforms and OpenAI.
- Challenging the status quo – By offering a “total solution,” Samsung aims to erode SK Hynix’s HBM dominance and provide a manufacturing alternative to TSMC for advanced AI hardware.
Supply issues for memory are likely to persist for some time, and the result is that chip-makers and others are working on more direct partnerships to get the supply they need. Samsung and AMD have inked a memorandum of understanding, broadening their partnership around memory chip supplies for AI infrastructure. The joint announcement covers next-gen high-bandwidth memory, optimized server DRAM, and, interestingly, potential foundry cooperation where Samsung would actually manufacture future AMD chips.
The agreement
The heart of the MOU is Samsung supplying HBM4 chips for AMD’s upcoming Instinct MI455X AI accelerators, plus optimized DDR5 memory for AMD’s sixth-generation EPYC processors. These components are built to power next-gen AI systems that bring together AMD Instinct GPUs, EPYC CPUs, and rack-scale architectures like AMD’s Helios platform — basically the entire hardware stack required to run large-scale AI workloads.
This deal isn’t all that surprising. Samsung already serves as a primary HBM supplier to AMD, providing the HBM3E chips inside AMD’s MI350X and MI355X accelerators today. What the new agreement really does is deepen and formalize something that was already well underway, pushing it forward into the next generation of memory technology.
The foundry piece is perhaps a little more interesting though. The two companies will explore Samsung serving as a contract manufacturer for future AMD chips, which would mark a significant leap beyond their current memory supply arrangement. It could position Samsung as a manufacturing alternative — or at least a complement — to TSMC, which fabricates the overwhelming majority of AMD’s processors right now.
The broader picture
The deal is one of two involving Samsung — Samsung and AMD went public with this deal just one day after Nvidia revealed its own foundry partnership with Samsung, with Jensen Huang announcing at GTC that Samsung is producing Nvidia’s latest AI chips on 4-nanometer technology. Industry analysts see this cluster of deals as part of an intensifying fight among semiconductor manufacturers to secure long-term advanced memory supply agreements, especially as HBM component shortages are putting serious demand pressure on the entire AI infrastructure market.
Samsung co-CEO Jun Young-hyun said the company is working with major customers to shift toward multi-year contracts spanning three to five years, and Samsung leadership described the chip industry as entering an “unprecedented supercycle” fueled by surging AI data center investment. The market clearly bought the narrative — Samsung shares jumped more than 7% on the news.
This partnership also challenges SK Hynix’s dominance in HBM technology. Samsung is trying to position itself as a comprehensive AI infrastructure supplier, with executives emphasizing Samsung as “the only company that can provide a ‘total solution’ connecting and supplying various core components inside AI servers.” That’s a bold claim. SK Hynix still holds a commanding lead in HBM market share and maintains its own deep ties to Nvidia and other major chip designers.
AMD’s AI strategy
AMD is going head-to-head with Nvidia in the AI accelerator market, and while Nvidia is the clear dominant player, AMD has been building steam. Just last month, AMD announced a massive agreement to potentially sell up to $6100 billion in AI processors to Meta Platforms over five years, with Meta able to purchase up to 10% in AMD’s stock. AMD previously struck a comparable deal with OpenAI. Fulfilling commitments at that scale requires a reliable, diversified memory supply chain — which makes this Samsung MOU as much about operational capacity as it is about the underlying technology.
The AI memory partnership isn’t the only front where the two companies are deepening ties. Earlier in March, Samsung and AMD announced broader strategic collaboration centered on telecommunications infrastructure. The companies demonstrated AI-RAN (Radio Access Network) breakthroughs, progressing from joint verification all the way to commercial deployments. Samsung’s AI-powered vRAN now runs on AMD EPYC processors for commercial-grade performance without needing additional accelerators, and the pair unveiled a “Network in a Server” (NIS) solution aimed at edge-AI enterprise applications like video analysis and sensor detection services.
