Samsung Electronics is seeking to expand its relationship with AMD beyond high-bandwidth memory, as the two semiconductor companies continue exploring a potential foundry partnership amid growing demand for artificial intelligence hardware.
AMD CEO Lisa Su reaffirmed the discussions during her October 7, 2026, visit to South Korea, where securing advanced memory supplies remained a central priority.
The negotiations follow a strategic agreement signed in March that positioned Samsung as a primary HBM4 memory supplier for AMD’s next-generation Instinct MI455X AI accelerators.
Industry reports suggest Samsung could use its increasingly valuable memory supply position to strengthen its negotiating power and potentially secure advanced chip manufacturing orders from AMD.
However, neither company has publicly confirmed a binding agreement to manufacture AMD’s advanced processors at Samsung Foundry.
The distinction matters because the outcome could affect manufacturing competition with TSMC and the supply chain supporting some of the world’s largest AI infrastructure projects.
Key Takeaways
- Samsung and AMD confirmed an HBM4 supply agreement in March 2026, targeting AMD Instinct MI455X accelerators.
- Lisa Su continues exploring Samsung Foundry cooperation, but no new advanced-node manufacturing contract has been publicly announced.
- Industry reports suggest Samsung may use its memory supply advantage as negotiating leverage, although the precise terms remain undisclosed.
- AMD’s Helios AI platform incorporates up to 31TB of HBM4 memory per rack, highlighting the importance of dependable memory supplies.
- Samsung does not hold an HBM4 monopoly: SK Hynix and Micron also compete in advanced high-bandwidth memory.
Samsung and AMD’s HBM4 Partnership Is Already Official
Samsung and AMD formally expanded their semiconductor relationship on March 18, 2026, through a memorandum of understanding covering advanced memory and computing technologies.
According to the official Samsung and AMD announcement, the companies agreed to coordinate primary HBM4 supply for the AMD Instinct MI455X accelerator.
The collaboration also includes next-generation DDR5 memory optimized for sixth-generation AMD EPYC processors, known by the codename Venice.
Samsung has described its broader capabilities in memory manufacturing, semiconductor fabrication, and advanced packaging as potential advantages for AMD’s future products.
More importantly, the March announcement explicitly confirmed that the companies would explore a foundry partnership.
That possibility resurfaced during Lisa Su’s October visit to Seoul.
According to Reuters’ October 7 report, Su indicated that AMD continued evaluating potential cooperation with Samsung across memory and chip manufacturing.
This establishes ongoing discussions, not an agreed transfer of advanced processor production.
Why HBM4 Memory Is So Important for AMD’s AI Strategy
High-bandwidth memory, commonly called HBM, is designed to move enormous amounts of data between memory and processors.
Unlike conventional desktop RAM, HBM uses vertically stacked memory chips positioned close to high-performance computing processors.
This arrangement delivers extremely high memory bandwidth while reducing the energy required to transfer data.
For AI accelerators, bandwidth is critical because modern models must constantly move substantial quantities of information during training and inference.
Samsung began commercial HBM4 shipments in February 2026.
Its official HBM4 announcement describes a design combining advanced DRAM manufacturing with a 4nm logic base die.
The company reports data transfer speeds of 11.7Gbps, with its technology capable of reaching up to 13Gbps.
AMD’s Instinct MI455X illustrates why this technology matters.
According to AMD’s official accelerator specifications, each MI455X incorporates 432GB of HBM4 memory and offers up to 23.3TB/s of theoretical memory bandwidth.
A complete AMD Helios rack combines 72 accelerators with approximately 31TB of HBM4 memory.
These are AMD’s published product specifications, rather than independently measured application benchmarks.
The larger implication is clear: AMD’s ambitions in AI infrastructure depend not only on powerful GPU designs but also on securing sufficient advanced memory and packaging capacity.
Does Samsung Really Have an HBM4 Monopoly?
Reports describing Samsung as having a monopoly over AMD’s HBM4 supply require an important qualification.
Samsung is a major HBM manufacturer, but it is not the industry’s only supplier.
SK Hynix and Micron also manufacture advanced high-bandwidth memory.
In fact, Micron’s official HBM product documentation confirms that its HBM4 products are in high-volume production.
The relevant issue is therefore not an industry-wide monopoly, but the availability of qualified memory capacity for specific customers.
AI chipmakers cannot necessarily switch suppliers immediately. Memory products must meet demanding requirements for performance, reliability, packaging, and integration with accelerator designs.
Even when multiple manufacturers exist, available production capacity may already be committed to other customers.
That constraint can increase the negotiating strength of suppliers capable of delivering the required products at scale.
In March, semiconductor industry publication TrendForce cited Korean reporting suggesting Samsung might seek additional foundry business by leveraging its HBM supply relationship with AMD.
However, the report did not disclose contractual terms proving that memory shipments were conditional on AMD transferring processor orders.
This remains a reported negotiating possibility rather than an established contractual fact.
Why Samsung Wants AMD’s Advanced Chip Manufacturing Business
Samsung’s ambitions extend beyond supplying memory.
The company operates a semiconductor foundry business capable of manufacturing chips designed by external customers.
Its principal competitor in advanced contract chip manufacturing is TSMC, which already produces many of AMD’s most sophisticated processors.
Securing a meaningful AMD manufacturing contract could help Samsung expand its advanced-node customer base.
Industry reports have pointed to Samsung’s second-generation 2nm process, known as SF2P, as one possible technology for future cooperation.
However, neither company has officially confirmed that an AMD product will be manufactured using SF2P.
Several commercial and technical questions remain.
Advanced processors require manufacturing processes capable of delivering suitable performance, power efficiency, reliability, and production yields.
Yield refers to the proportion of manufactured chips that meet the required specifications.
A process with insufficient yield can make a complex chip expensive or commercially impractical, even when its underlying technology appears promising.
For AMD, adding another advanced manufacturing partner could eventually provide supply-chain flexibility.
But changing foundries is not as straightforward as ordering the same chip from another factory.
Semiconductor designs often require extensive engineering work to adapt to different manufacturing processes.
That makes any potential cooperation a long-term strategic decision rather than a simple short-term response to memory shortages.
What Happens Next for Samsung, AMD, and TSMC?
Three developments will determine whether the current discussions translate into a significant manufacturing agreement.
First, Samsung must continue demonstrating that it can supply the advanced memory required for AMD’s expanding AI accelerator portfolio.
Second, its foundry business must offer competitive manufacturing technology and adequate production capacity for the designs AMD might consider outsourcing.
Third, AMD must determine whether diversifying part of its advanced chip manufacturing would provide enough benefits to justify the additional engineering and commercial complexity.
For now, the verified position remains straightforward: Samsung and AMD have an established memory relationship, and both companies are exploring potential foundry cooperation.
Reports that Samsung is using HBM4 supply as leverage are plausible within the broader industry context, but they should not be presented as proof of coercion or a completed manufacturing deal.
The wider story is about changing negotiating power in the AI semiconductor industry.
Advanced memory, processor manufacturing, and chip packaging are becoming increasingly interconnected as data center systems grow more complex.
Companies capable of supplying several of these technologies may gain opportunities to develop broader customer relationships.
Samsung’s HBM4 partnership with AMD could become one such opportunity, but the decisive evidence will be an officially disclosed foundry agreement and actual manufacturing commitments.
Until then, the negotiations illustrate both the importance of reliable HBM4 supply and the growing competition to manufacture the next generation of AI processors.
For more reporting on AI hardware, processors, and semiconductor industry developments, explore the latest technology coverage at Tech News Home.
Frequently Asked Questions About Samsung, AMD, and HBM4
Is Samsung forcing AMD to use its foundry services?
There is no publicly confirmed evidence that Samsung has forced AMD to transfer advanced processor production. Industry reporting suggests Samsung’s HBM4 supply relationship may provide negotiating leverage, but neither company has announced a binding advanced-node foundry agreement.
Why does AMD need Samsung’s HBM4 memory?
AMD uses HBM4 in its next-generation Instinct MI455X accelerators and Helios AI infrastructure. The technology provides high memory bandwidth needed for demanding AI training and inference workloads.
Will Samsung manufacture future AMD CPUs and GPUs?
Samsung and AMD have acknowledged discussions about possible foundry cooperation. However, they have not publicly confirmed which future processors, if any, would be manufactured using Samsung’s advanced foundry processes.
Official Sources and Further Reading
- Samsung — Strategic HBM4 Collaboration With AMD — Official March 18, 2026 announcement confirming HBM4 cooperation, DDR5 development, and exploratory foundry discussions.
- Reuters — AMD Continues Exploring Samsung Foundry Partnership — October 7, 2026 reporting on Lisa Su’s statements regarding additional cooperation.
- AMD — Instinct MI400 Series and MI455X Specifications — Official technical information about HBM4 capacity, memory bandwidth, and the AMD Helios architecture.
- Samsung — Commercial HBM4 Production Announcement — Technical details and manufacturing information about Samsung’s HBM4 products.
- TrendForce — Reported HBM4 Negotiating Leverage — Industry reporting on the possibility that memory supply discussions could influence foundry negotiations.
- Micron — Official HBM4 Product Information — Independent manufacturer documentation confirming HBM4 production outside Samsung.