AMD is reportedly advancing discussions with Samsung Electronics to manufacture some of its upcoming chips using the Korean group’s 2-nanometer process. This information, published by the Korean media outlet EDaily and picked up by various industry publications, comes at a time when demand for CPUs, GPUs, and AI accelerators is straining the capacity of major foundries.
The operation has not been officially confirmed by AMD or Samsung, so it should be viewed as an ongoing negotiation rather than a finalized contract. Nevertheless, it aligns with an increasingly visible trend: leading chip designers can no longer rely on a single advanced manufacturing partner if they want to sustain the growth pace required by AI.
According to sources in South Korea, the talks accelerated after AMD CEO Lisa Su’s visit to Korea in March, during which she toured Samsung’s facilities in Pyeongtaek. That same month, AMD and Samsung signed a memorandum to strengthen their collaboration in advanced memory for AI, including HBM4 for future Instinct GPUs and DRAM solutions for the sixth-generation EPYC processors, codenamed Venice.
AI is also increasing pressure on CPUs
Over the past two years, the conversation surrounding AI infrastructure has been dominated by GPUs. NVIDIA has set the pace, AMD has bolstered its Instinct family, and hyperscalers have competed to secure accelerators, HBM memory, high-speed networks, and electrical capacity. But as AI agent systems gain prominence, CPUs are regaining importance.
An AI system running agents doesn’t just process isolated requests. It coordinates workflows, calls tools, moves data, manages services, handles queues, connects to databases, orchestrates accelerators, and maintains processes over extended periods. GPUs are essential for intensive computation, but CPUs remain the general-purpose brain that organizes much of the platform.
AMD’s results reflect this trend. The company closed Q1 2026 with $10.253 billion in revenue, a 38% increase year-over-year. Its data center division reached $5.8 billion, up 57%, driven by strong demand for EPYC processors and growth in Instinct GPUs. Lisa Su stated that inference and AI agent systems are boosting demand for high-performance CPUs and accelerators.
This growth also has an uncomfortable side: increased manufacturing capacity is needed. AMD’s recent success has largely relied on TSMC, both for CPUs and advanced GPUs. The case of Venice EPYC products is especially illustrative. In April 2025, AMD announced Venice as the first HPC product fabricated on TSMC’s N2 2nm node, solidifying its partnership. However, this also highlighted the heavy dependence of upcoming products on in-demand advanced nodes.
Samsung aims to regain its status as a serious foundry alternative
Samsung Foundry has been working for years to rebuild credibility against TSMC in advanced manufacturing. The company aggressively adopted Gate-All-Around transistors at 3nm but faced doubts regarding performance, customer adoption, and process efficiency. The move to 2nm is an opportunity to reposition itself—provided it can improve yields, ecosystem integration, capacity, and confidence among major clients.
Samsung presents its SF2 process as an evolution of its GAA technology, with improvements in performance, power consumption, and area over previous generations. It has also worked on compatibility with SF3, IP readiness, high-density libraries, and solutions for HPC, a market where rapid interfaces, advanced memory, and packaging are as crucial as the manufacturing node.
The potential entry of AMD into Samsung’s foundry could be symbolically significant. Samsung already supplies chips for Tesla, including AI5 and AI6, and its Taylor, Texas plant is preparing to manufacture AI chips using 2nm processes. Gaining a share in AMD’s future production would secure a major cloud data center customer for Samsung and bolster its position as a viable alternative to TSMC for AI workloads.
The AMD-Samsung relationship is also not starting from zero. The announced March agreement covers HBM4 memory for the AMD Instinct MI455X GPU and advanced DRAM for EPYC Venice and AI platforms like AMD Helios. Reuters reported at the time that the two companies were exploring potential foundry collaboration, providing additional context to currently published reports from Korea.
Diversification doesn’t mean breaking from TSMC
Producing 2nm chips at Samsung should not be interpreted as AMD breaking away from TSMC. Instead, it’s likely a strategy of dual sourcing. TSMC remains the leader in advanced nodes and packaging for HPC, and AMD has publicly collaborated with TSMC on N2. However, TSMC’s capacity is under enormous pressure from Apple, NVIDIA, AMD, Qualcomm, MediaTek, Tesla, and major AI clients.
In this context, having a second source can be a strategic advantage—both for pricing and scheduling. If part of AMD’s CPU or AI chip supply can be produced at Samsung, the company can better respond to demand spikes, negotiate with more flexibility, and reduce concentration risks. For complex products, duplicating foundry sourcing isn’t trivial—it requires adapting designs, libraries, validation, packaging, and performance tuning. But the benefits can outweigh the challenges if expected volumes are high enough.
Samsung’s challenge will be demonstrating its ability to produce high-performance chips at the quality level AMD requires. In server CPUs, the margin for error is small; customers demand performance, efficiency, reliability, compatibility, and availability over years. Yield issues or power consumption problems could impact costs, timelines, and market trust.
For AMD, the opportunity is clear. The company is entering a phase where its data center portfolio depends not only on EPYC but also on Instinct, Helios, networking, memory, and agreements with major cloud providers. If AI agent workloads become dominant, server CPU demand could grow alongside GPUs. Scaling this offering will require more wafers, more HBM, advanced packaging, and greater industrial flexibility.
The potential alliance with Samsung also has geopolitical implications. Manufacturing part of the next-generation chips in South Korea or at the Taylor, Texas facility would diversify geographically from Taiwan—adding resilience from a geopolitical perspective, which is meaningful for U.S. and European clients concerned about supply chain stability.
Of course, the most crucial details remain to be confirmed: official confirmation, production volumes, specific products, and timelines. Until AMD and Samsung clarify whether a contract exists, what chips will be produced, and where, this news should be viewed as a market signal. A strong one, at that: AI demand is pushing leading semiconductor designers to seek more advanced capacity, and Samsung aims to capitalize on this window to re-enter the global foundry front line.
Frequently Asked Questions
Has AMD confirmed it will manufacture 2nm chips at Samsung?
No. Currently, these are advanced discussions reported by Korean media and industry sources. AMD and Samsung have not announced an official 2nm manufacturing contract.
Why would AMD seek Samsung’s services if it already works with TSMC?
To diversify manufacturing capacity, reduce dependence on a single foundry, and better respond to the high demand for CPUs and AI chips.
What role does EPYC Venice play in this story?
EPYC Venice is AMD’s next-gen server CPU line, and it has already been announced as the first HPC product fabricated on TSMC’s N2 2nm node.
What would Samsung gain from an agreement with AMD?
Samsung Foundry would bolster its credibility in advanced nodes, gain a key data center customer, and strengthen its position against TSMC in AI and HPC chips.