Agility Robotics Goes Public in $2.5B SPAC Deal, Becoming First US-Listed Pure-Play Humanoid MakerGoogle Delays Gemini 3.5 Pro to July 17 for Ground-Up Architectural RebuildICML 2026 Wraps in Seoul: Record 24,000 Submissions, Agentic AI Everywhere, and a Peer-Review ReckoningTwelve Deals in Eighteen Months: Why Ant Group Is Buying Into the Robot FutureFrom 'Best Model Wins' to 'Best Fit Wins': The Frontier AI Glut Is Rewriting the RaceHumanoid Robots Hit the Public Markets — and the Hype Meets Its First Earnings TestSmart Glasses Maker Even Realities Hits $1B Valuation With $150M Round Led by Meituan and TencentTripo AI Raises $150M Series A3 as 3D Foundation Models Push Toward Interactive WorldsAgility Robotics Goes Public in $2.5B SPAC Deal, Becoming First US-Listed Pure-Play Humanoid MakerGoogle Delays Gemini 3.5 Pro to July 17 for Ground-Up Architectural RebuildICML 2026 Wraps in Seoul: Record 24,000 Submissions, Agentic AI Everywhere, and a Peer-Review ReckoningTwelve Deals in Eighteen Months: Why Ant Group Is Buying Into the Robot FutureFrom 'Best Model Wins' to 'Best Fit Wins': The Frontier AI Glut Is Rewriting the RaceHumanoid Robots Hit the Public Markets — and the Hype Meets Its First Earnings TestSmart Glasses Maker Even Realities Hits $1B Valuation With $150M Round Led by Meituan and TencentTripo AI Raises $150M Series A3 as 3D Foundation Models Push Toward Interactive WorldsAgility Robotics Goes Public in $2.5B SPAC Deal, Becoming First US-Listed Pure-Play Humanoid MakerGoogle Delays Gemini 3.5 Pro to July 17 for Ground-Up Architectural RebuildICML 2026 Wraps in Seoul: Record 24,000 Submissions, Agentic AI Everywhere, and a Peer-Review ReckoningTwelve Deals in Eighteen Months: Why Ant Group Is Buying Into the Robot FutureFrom 'Best Model Wins' to 'Best Fit Wins': The Frontier AI Glut Is Rewriting the RaceHumanoid Robots Hit the Public Markets — and the Hype Meets Its First Earnings TestSmart Glasses Maker Even Realities Hits $1B Valuation With $150M Round Led by Meituan and TencentTripo AI Raises $150M Series A3 as 3D Foundation Models Push Toward Interactive Worlds
SK Hynix HBM4 high-bandwidth memory stack
TrendForce
News

NVIDIA Requests Industry-First 16-Hi HBM4 as SK Hynix, Samsung and Micron Race to Deliver

NVIDIA has asked memory makers for 16-high HBM4 stacks, an industry first, setting off a race between SK Hynix, Samsung and Micron to solve steep manufacturing hurdles by late 2026.

D
Daniel ParkAI Correspondent
3 min read

NVIDIA has asked its memory suppliers to prepare 16-high HBM4 stacks, an industry first, touching off a fresh competition between SK Hynix, Samsung and Micron to be first to mass-produce the most complex high-bandwidth memory built to date.

Why 16-high is so hard

Today's leading HBM4 products stack 12 DRAM dies. Moving to 16 layers packs more memory capacity and bandwidth into each stack, which AI accelerators need as models and context windows grow. But the added height sharply raises manufacturing difficulty. Each die must be thinned further so the full stack fits within height limits, pushing wafer thickness from around 50 microns toward roughly 30 microns.

Thinner wafers warp and crack more easily, and bonding 16 dies without defects strains packaging and yield. No supplier has commercialized 16-high HBM before, and the companies are still working through stacking, warpage and thermal challenges internally rather than under signed contracts.

A late-2026 target

NVIDIA is said to want the 16-high parts available around the fourth quarter of 2026, aligning with its next-generation accelerator roadmap. Concrete supply agreements have not yet been finalized, with discussions focused on initial production volumes and which vendors can hit the timeline.

The three memory makers already split HBM4 supply for NVIDIA's current platform, after all passed certification earlier this year. Supply-chain estimates put SK Hynix at roughly 60 to 70 percent of that volume, with Samsung around a quarter and Micron supplying the remainder. The 16-high round offers each a chance to reset that balance.

Stakes for the memory market

HBM has become one of the most lucrative and closely watched segments in semiconductors, and leadership in the newest tier carries outsized weight for pricing and market share. For the Korean giants in particular, being first to deliver reliable 16-high HBM4 would reinforce their position at the center of the AI hardware supply chain, while any stumble risks ceding ground to rivals racing on the same aggressive timeline.

Newsletter

Get Lanceum in your inbox

Weekly insights on AI and technology in Asia.

Share

More in News

Lanceum

Independent coverage of AI and technology across Asia. We go beyond headlines to explain what matters.

Colophon

Typeset in Space Grotesk & DM Serif Display. Built with Nuxt & Tailwind. Powered by curiosity.

© 2026 Lanceum. All rights reserved.

Independent • Rigorous • Asia-Focused