Overview
This report examines Intel Foundry's advanced packaging strategy, EMIB, ZAM, and XBM, as a challenger to TSMC's CoWoS in AI accelerator packaging. HBM has become the default memory solution for AI accelerators, but its reliance on massive silicon interposers drives up cost, defect risk, and capacity constraints. The report assesses whether Intel's three-stage roadmap, with EMIB today, ZAM as the mid-decade bridge, and XBM as the long-term reinvention, can realistically challenge TSMC's position.
What the Full Report Will Cover
The report breaks down EMIB, ZAM, and XBM individually, covering how each works and what it claims over CoWoS/HBM, and maps Intel's likely rollout sequence (EMIB now, ZAM around 2028 to 2030, XBM 2030 and beyond), including known and rumored design-ins. It then evaluates the key technical and ecosystem hurdles (thermal budget, SerDes latency, HBM/JEDEC lock-in, foundry execution) and closes with a hypothesis on whether the three technologies could eventually converge into a single architecture.
Key Questions the Report Answers
- Why is HBM combined with CoWoS the most expensive, constrained part of AI accelerator packaging today?
- What are EMIB, ZAM, and XBM, and what do they claim to improve over CoWoS/HBM?
- What is Intel's realistic timeline, and who is already showing adoption interest?
- What technical and ecosystem barriers stand between Intel's roadmap and real market share?
- Could EMIB, ZAM, and XBM converge into one architecture, and what would have to go right for that to happen?
