NVIDIA Meets Samsung and SK Hynix as HBM Supply Negotiations Intensify

The Quick Take

• NVIDIA executives met with Samsung and SK Hynix representatives to discuss HBM supply chain capacity for AI accelerator chips
• The talks centre on securing High Bandwidth Memory allocations for NVIDIA's Blackwell Ultra and successor GPU architectures through 2027
• SK Hynix currently supplies approximately 70% of NVIDIA's HBM3E demand while Samsung works to qualify its HBM4 product
• Supply chain negotiations reflect NVIDIA's drive to diversify memory sourcing and reduce dependence on a single supplier

NVIDIA's supply chain discussions with Samsung and SK Hynix cut to the heart of the AI infrastructure buildout's critical constraint: High Bandwidth Memory capacity. HBM is the bandwidth-intensive memory architecture embedded in AI training accelerators, and qualified HBM supply has consistently been the binding limitation on NVIDIA's ability to meet hyperscaler GPU demand since 2023. With next-generation Blackwell Ultra and future architectures requiring HBM4 at even higher bandwidth specifications, qualifying Samsung as a credible second source is strategically essential for production resilience.

The market implications split between the two Korean suppliers. SK Hynix benefits from incumbent status—its HBM3E product has already passed NVIDIA qualification and commands premium ASPs—but faces a credible threat to margin if Samsung qualifies HBM4 on NVIDIA's timeline. For Samsung, HBM qualification with NVIDIA represents a major revenue catalyst: estimates suggest NVIDIA-qualified HBM contracts could be worth $8–12 billion annually at peak volume. The competitive dynamic is also a potential margin tailwind for the HBM ecosystem as NVIDIA's preference for supply certainty may support above-market pricing through 2026.

Qualification milestones to watch: Samsung has targeted HBM4 mass production for H1 2027, and NVIDIA design-win announcements ahead of that date would reprice Samsung semiconductor equities meaningfully. SK Hynix's Q4 2026 guidance will include HBM capacity utilisation and NVIDIA-specific volume commentary framing the competitive landscape. Broader AI semiconductor supply chain investors should monitor TSMC's CoWoS advanced packaging capacity as a parallel bottleneck—NVIDIA GPU production is jointly constrained by HBM allocation and advanced packaging throughput.

Synthesized from 1 source — full coverage, sentiment breakdown, and forward signals below.