AMD and Samsung expand HBM4 partnership for next-generation AI accelerators

AMD and Samsung have expanded their partnership on HBM4 — the next generation of high-speed memory for AI accelerators. The new memorandum was signed in Pyeongtaek, where Samsung has one of its largest memory production complexes.

What they agreed on

Rumors about AMD's CEO visiting South Korea have now been officially confirmed: the company announced the signing of a new memorandum of understanding with Samsung Electronics.

The document formalizes an expanded collaboration around HBM4 — memory that will become critically important for future AI accelerators.

For now, this is not about launching a specific product, but about closer joint work at an early stage, when memory requirements, compatibility, and manufacturing plans are being defined.

The signing location itself is also telling. The parties met in Pyeongtaek, where Samsung has a major memory production cluster.

For AMD, this is not just a symbolic gesture: the AI hardware market has long been constrained not only by compute power, but also by access to advanced memory stacks.

The faster models and workloads grow, the more important it becomes for accelerator makers to reserve technology supply chains in advance and establish direct coordination with HBM suppliers.

Why HBM4 matters

HBM is a stacked memory technology with very high bandwidth that is placed next to the compute die and helps the accelerator move data faster.

For training and inference tasks involving large models, this is one of the key components: if memory falls short on speed, capacity, or energy efficiency, the overall gains from a more powerful chip quickly hit a ceiling.

That is why the move to HBM4 matters not on its own, but as part of the next step in the development of server AI platforms.

The expanded cooperation between AMD and Samsung could affect several practical areas at once:

• earlier tuning of future AMD accelerators for the new generation of memory
• joint validation of thermal, power, and signal characteristics
• more predictable supply planning for data centers and OEM partners
• lower risks associated with HBM shortages during peak demand periods

Amid explosive demand for AI infrastructure, HBM has become one of the scarcest elements in the entire supply chain.

Competition is no longer only about the best GPU or accelerator, but also about access to memory, packaging, and production slots.

That is why agreements like this often play a strategic role long before finished cards and servers reach the market.

What the market gets

For Samsung, this is a chance to strengthen its position in the most sensitive segment of memory for AI.

The company has long competed for HBM orders, and deeper ties with AMD increase its chances of being a visible player in the next generation of accelerators.

For AMD, the benefit is different: the broader and deeper its network of technology partners, the more resilient its roadmap becomes.

In a segment where a delay in memory can derail the launch of an entire accelerator lineup, that kind of hedge is especially important.

It is also important that the memorandum is not a final contract for mass shipments, but a framework for joint work.

But these are exactly the kinds of documents that usually show where companies plan to concentrate engineering resources over the coming quarters.

For AMD customers, this is a signal that the company is laying the groundwork for the next generation of AI accelerators in advance and does not want to remain dependent on a single supply scenario at a time when the market is demanding ever more compute power.

What it means

The AI accelerator market is increasingly dependent not only on chip architecture, but also on access to memory.

The expansion of the AMD-Samsung partnership shows that the fight for AI leadership is now taking place at the HBM4 level as well, where the winner is not simply the one with the faster processor, but the one that assembles a resilient supply chain earlier.