Samsung will provide next-generation high-bandwidth memory, HBM4, as the primary supplier for AMD's MI455X accelerators used in corporate data centres, the two companies announced this week. The move carries implications that extend well beyond commercial memory contracts into questions of semiconductor supply chain resilience and alliance capabilities in a period of acute geopolitical competition.
The signing ceremony was held at Samsung's most advanced chip manufacturing complex in Pyeongtaek, Korea, attended by Dr. Lisa Su, Chair and CEO of AMD, and Young Hyun Jun, Vice Chairman and CEO of Samsung Electronics. The memorandum of understanding formalises Samsung's position as a critical supplier at a moment when TSMC's advanced-node capacity is about three times short of major customer requirements.
The technical specifications underscore Samsung's technical progress. Samsung's HBM4 is built on its most advanced 6th-generation 10-nanometer class DRAM process and a 4nm logic base die, featuring processing speeds of up to 13 gigabits-per-second and maximum 3.3 terabytes-per-second bandwidth that exceeds industry standards. For comparison, Samsung is the third supplier to secure approval from Nvidia, following SK Hynix and Micron, indicating hard-won validation from the most demanding customers in the AI supply chain.
Yet Samsung's path to this agreement has been neither smooth nor assured. Samsung Electronics is currently lagging behind its crosstown rival SK Hynix in the HBM race, having failed to secure supply deals for its advanced memory chips with Nvidia, while SK Hynix is currently the primary provider of the latest HBM chips to Nvidia. Samsung holds about a 22% share of the global HBM market, compared with market leader SK Hynix's 57%. The AMD deal represents a significant win in Samsung's efforts to narrow that gap.
Beyond memory supply, the two companies will also discuss opportunities for foundry partnership, through which Samsung would provide foundry services for next-generation AMD products. This element of the agreement points toward a structural shift in how major AI chipmakers manage manufacturing risk.
Global technology firms are increasingly shifting toward Samsung Foundry as Taiwan Semiconductor Manufacturing Company (TSMC) struggles to meet skyrocketing demand for advanced chips driven by artificial intelligence expansion. AMD has primarily outsourced AI chip production through TSMC, but surging semiconductor demand has pushed TSMC's advanced process capacity to its limits, creating the need for new partners. With TSMC's 3nm and 5nm capacities booked through 2026, Samsung's more competitive wafer prices between $22,000 and $25,000 are fuelling a broader industry shift to diversify chip supply chains.
The strategic implications warrant serious attention. For over a decade, the semiconductor industry has concentrated manufacturing at a single point of geographic and commercial failure: Taiwan. TSMC is the undisputed leader in pure-play foundry manufacturing, having built deep trust with fabless companies such as Nvidia, AMD, Apple, and Qualcomm by focusing exclusively on manufacturing and avoiding competition with its customers in chip design. That dominance has delivered genuine benefits in process technology leadership and manufacturing scale. It has also created a structural vulnerability.
Most AI accelerators from hyperscalers and GPU vendors are manufactured at TSMC, which also dominates CoWoS packaging capacity, giving TSMC structural leverage in the AI supply chain. In an environment where the memory industry is undergoing its most significant transformation in decades, driven by a desperate need for High-Bandwidth Memory to power the next generation of AI supercomputers, with major manufacturers like SK Hynix and Samsung aggressively reallocating up to 40% of their advanced wafer capacity toward specialised AI memory, capacity constraints are no longer theoretical.
Samsung's willingness to absorb foundry work at competitive pricing reflects both capability and opportunity. Samsung pioneered early gate-all-around transistors at 3nm and targets 2nm leadership, and is expanding U.S. manufacturing to attract Western customers seeking geographic diversification. The Pyeongtaek agreement signals that major chipmakers are now willing to use Samsung not because they have lost faith in TSMC, but because waiting for unavailable capacity amounts to accepting strategic delay. In AI markets, delay is defeat.
The deal also reflects a geopolitical calculation. Samsung's presence provides customers with an important second source at advanced nodes. As tensions over Taiwan intensify and governments worldwide push for domestic semiconductor manufacturing, having alternatives to a single Taiwan-based supplier has moved from luxury to necessity.
The challenge for Samsung remains real. Samsung has historically faced yield consistency challenges compared with TSMC at advanced nodes. Success in the foundry business requires not just competitive pricing or available capacity, but consistent delivery of working silicon at scale. That test still lies ahead. The AMD memory deal is a win; the foundry partnership—should it materialise at volume—would represent something more significant: proof that semiconductor supply chains need not concentrate all advanced manufacturing in a single geography or company.