Samsung and Nvidia Discuss HBM4E and HBM5 Chip Cooperation

Key Takeaways

• Samsung and Nvidia are expanding their partnership beyond current AI accelerator and autonomous driving chips
• The companies discussed long-term cooperation on HBM4E and HBM5 memory chips
• Samsung shipped HBM4E samples within 3 months of starting HBM4 mass production

What Happened in Seoul

Samsung Electronics' co-CEO and chip division head Jun Young-hyun met with Nvidia CEO Jensen Huang in Seoul on Monday. The two executives discussed cooperation on next-generation foundry chips, extending a partnership that already covers autonomous driving and AI accelerator chips.

Jun described the discussions as extensive, covering long-term cooperation on high-bandwidth memory products. The talks specifically addressed HBM4E and HBM5 chips, the next generations of memory technology that will power future AI data centers.

“The best conversation I have ever had with Jensen Huang.”

— Jun Young-hyun, Head of Samsung Electronics' Device Solutions Division

Why HBM Matters for AI

High-bandwidth memory is the critical link between AI processors and the data they need to crunch. As AI models grow larger, they demand faster memory with higher bandwidth. HBM stacks memory chips vertically and connects them with thousands of tiny wires, delivering data speeds that traditional memory cannot match.

Samsung's newly sampled HBM4E chips offer data transfer performance of 16 Gbps. That represents a significant leap that could help Nvidia's future AI accelerators handle even larger models. The company shipped these samples within 3 months of starting HBM4 mass production, a rapid turnaround that signals Samsung's urgency to compete in this space.

Samsung's Position in the HBM Race

Samsung is playing catch-up in the HBM market. SK Hynix currently leads, having secured most of Nvidia's HBM orders for its H100 and subsequent AI chips. This meeting signals Samsung's push to reclaim market share by positioning itself as a collaborative partner rather than just a supplier.

The discussions covered 8 generations of memory roadmap, from HBM4 through HBM5. That depth of planning suggests both companies see this as more than a transactional relationship. Samsung wants to embed itself in Nvidia's multi-year AI accelerator roadmap.

“Samsung is a vital partner for the future of AI infrastructure as we push toward the next generation of computing.”

— Jensen Huang, CEO of Nvidia

The Foundry Angle

The meeting covered more than memory. Jun specifically mentioned next-generation foundry chips, pointing to Samsung's ambitions to manufacture more of Nvidia's silicon. Samsung's foundry business has struggled against TSMC, which produces most of Nvidia's current GPU lineup.

One technical approach under discussion involves using Samsung's 4nm process for the base die of HBM4 memory modules. This integration could differentiate Samsung's offering from competitors. Online communities tracking semiconductor developments have flagged this as a potential competitive advantage, though real-world benchmarks will determine if the approach delivers.

What This Means Going Forward

For Samsung, this meeting represents a strategic pivot. The company is moving from a traditional supplier role toward joint development partnerships. That shift matters because AI chip makers increasingly want memory partners who can co-develop products, not just fulfill orders.

For Nvidia, diversifying its memory supply chain reduces risk. Relying too heavily on any single supplier creates vulnerability. Having Samsung as a capable alternative to SK Hynix gives Nvidia negotiating leverage and supply security.

The companies already work together on autonomous driving chips and AI accelerators. Expanding to HBM and foundry cooperation could deepen that relationship significantly over the next several years.

Frequently Asked Questions

What is HBM4E?

HBM4E is the enhanced version of fourth-generation high-bandwidth memory. Samsung's samples deliver 16 Gbps data transfer speeds, designed for next-generation AI accelerators.

Why did Jensen Huang meet with Samsung's chip chief?

The two executives discussed expanding their existing partnership to include HBM4E, HBM5, and next-generation foundry chips for AI infrastructure.

How does Samsung compare to SK Hynix in HBM?

SK Hynix currently leads the HBM market and supplies most of Nvidia's memory. Samsung is working to catch up through faster product development and closer partnerships with customers like Nvidia.

What chips do Samsung and Nvidia already work on together?

The companies currently collaborate on autonomous driving chips and AI accelerator chips, with the new discussions extending to memory and foundry cooperation.

Source: Tech-Economic Times / ET