2 September 2026

Experimental memory tech could speed up AI model training

• Magnonics and vertical FeRAM are early-stage memory technologies that researchers say could offer faster access speeds than current high-bandwidth memory solutions.
• If commercialized, these technologies could store more data in less physical space while maintaining speed, addressing a major constraint in training large AI models.
• Developing these technologies requires substantial funding and experienced teams, making commercialization uncertain despite the potential benefits.

How it was covered

TLDR AITLDR editorial team

Early-stage memory technologies like magnonics and vertical FeRAM could yield faster access speeds than current HBM or HBM-bandwidth with NAND-like density. The article argues these would be revolutionary if commercialized, though they require founding teams capable of raising hundreds of millions to billions of dollars.