AMD and Samsung's HBM4 Deal: What's Signed, What's Still Pending, and Why It Matters

Ties to AMD's Helios AI Platform

Despite the pending formal contract, the partnership is already operational. Samsung HBM4 memory is powering AMD's Helios AI server rack system, which entered production in mid-2026 . At the heart of Helios is the AMD Instinct MI455X GPU, equipped with 12-high HBM4 stacks delivering 432 GB of capacity and 23.3 TB/s of bandwidth per chip — roughly 50% more memory per GPU than the prior generation .

Samsung's principal engineer confirmed at AMD's Advancing AI 2026 event that Samsung HBM4 now ships inside the MI455X GPUs and Helios racks . Shipments of Helios are scheduled to begin in late Q3 2026, ramping into early 2027 . A Helios rack combining up to 72 MI455X units delivers a total of 31 TB of HBM4 capacity and 1.7 PB/s of memory bandwidth .

AMD's Customer Agreements Create Huge Downstream Demand

AMD has secured major AI chip customers, creating enormous downstream demand that flows back to Samsung's HBM4:

• OpenAI: Multi-year deal announced October 2025, valued at tens of billions in annual revenue. OpenAI receives an option to acquire up to ~10% of AMD, and the agreement includes 6 GW of Instinct GPU deployment .
• Meta: 5-year deal announced February 2026, valued at up to $60 billion in AI chips. Meta also gets an option for up to 10% of AMD .
• Anthropic: Strategic partnership plus a $5 billion investment, announced July 2026. Anthropic will deploy 2 GW of AMD Instinct MI450 GPUs in Helios racks, representing tens of billions in total server value .
• Microsoft: Won as a Helios customer; Microsoft is deploying AMD Helios racks for its AI infrastructure .
• Oracle: Previously announced as the first publicly available AI supercluster powered by MI450 GPUs .

One source notes that Anthropic was listed as AMD's highest-priority customer in internal code, alongside Meta .

Samsung's HBM4 Resurgence

Samsung has staged a notable recovery in the HBM market after struggling with HBM3E qualification delays in 2024–2025:

• Market share shift: In Q3 2025, SK Hynix held 53% of the HBM market, Samsung 35%, and Micron 11% . By late 2025/early 2026, Samsung began aggressively closing the gap. By mid-2026, market projections from TrendForce and Bernstein forecast Samsung's HBM share rising from 27% in 2025 to 37% in 2026, while SK Hynix's share was expected to drop from 56% to 43% .
• HBM4 first-mover advantage: Samsung shipped the industry's first commercial HBM4 in February 2026 and began mass production ahead of schedule . It sold out its entire 2026 HBM4 production capacity, signaling extraordinary demand .
• Capacity expansion: Samsung plans to boost total HBM production to ~250,000 wafers/month by end of 2026 — a ~50% increase from its prior capacity of ~170,000 wafers . The company also aims for HBM4 to account for more than half of its total HBM output in 2026 .
• Nvidia supply: Samsung started HBM4 production for Nvidia in early 2026, alongside its AMD supply .
• Profit recovery: Samsung Electronics tripled its profit in Q4 2025 as HBM revenues surged, with Counterpoint Research noting "Samsung is clearly back" .
• Competitive dynamics: Some analysts see Samsung gaining an early HBM4 edge by embracing yield risk more aggressively than SK Hynix, though SK Hynix still leads overall HBM with ~60% market share entering 2026 .

Summary

AMD and Samsung have a signed MoU and active HBM4 shipments flowing into the Helios platform, but the final binding volume contract is still being finalized as of late July 2026. The HBM4 link is strategically critical: AMD's massive customer agreements with OpenAI, Meta, Anthropic, Microsoft, and Oracle translate directly into Samsung HBM4 demand. Samsung, meanwhile, has made a strong comeback from earlier HBM3E quality issues — it was first to commercial HBM4, sold out its 2026 capacity, and is gaining share against SK Hynix in the next-generation memory race.