Piecemakers bets edge AI devices will diverge from reliance on HBM — custom-designed memory fuses DRAM stack directly to the processor using hybrid bonding
- ID
- 25148
- Status
- summarized
- Published
- 16 Sep 2026, 10:36 PM
- Fetched
- 17 Sep 2026, 12:43 AM
- Provider
- Tom's Hardware
- Category
- technology
- Original URL
- https://www.tomshardware.com/tech-industry/semiconductors/piecemakers-bets-edge-ai-devices-will-diverge-from-reliance-on-hbm-custom-designed-memory-fuses-dram-stack-directly-to-the-processor-using-hybrid-bonding
- Source URL
- https://www.tomshardware.com/feeds/all
Summary
- Score
- 5.0
- Created
- 17 Sep 2026, 12:44 AM
- Tags
- Audience
- developersai_ml_learnerssaas_startup_founders
What happened
Piecemakers, a Nanya-backed DRAM designer, debuted on Taiwan's Emerging Stock Board at NT$740, betting that edge AI inference will use DRAM stacked directly onto processors via hybrid bonding rather than HBM. The company positions its technology between Nvidia's SRAM-only Groq LPU and HBM, with stacked DRAM shipping in 2027. Currently, design fees—not chips—drive revenue, with AI custom-design work accounting for ~40% of H1 2026 revenue.
Why it matters
If you're building edge AI products, this signals a potential 2027 hardware path where inference doesn't require expensive HBM, which could lower per-device costs for on-device AI. But today this is a pre-revenue bet—design fees fund the company and the first volume customer program hasn't contributed financially yet—so it's a trend to watch, not a procurement decision to make now.
Discussion angle
Will edge AI inference actually diverge from HBM, or is this a niche play? Compare the cost/performance tradeoff of stacked DRAM-on-processor vs. HBM for Malaysian builders deploying on-device AI in IoT, automotive, or consumer devices.