Intel is hinting at a memory comeback as AI demand turns DRAM from a dull commodity into strategic gold dust.
Intel chief executive Lip-Bu Tan has suggested memory is no longer just a commodity business. Chipzilla now sees room for plenty of innovation.
For those not in the know, Intel began life in 1968 as a memory outfit andits early products included the 3101 SRAM and 1103 DRAM.
Chipzilla’s last big memory adventure was Optane, based on 3D XPoint technology co-developed with Micron. Optane was killed off in 2022 after failing to take off as Intel had hoped. Competing technologies such as HBM and multilayer 3D stacked memory looked cheaper and more promising.
Four years later, Chipzilla appears to be peering back through the DRAM door. This time, AI demand is doing the door-opening. Tan recently hired former SK Hynix chief executive Seok-Hee Lee as executive vice-president of foundry and advanced packaging technologies.
Chipzilla has been quietly working with partners on memory ideas such as XBM and ZAM. These are meant to address bottlenecks caused by AI’s endless hunger for bandwidth.
Older attempts, including Hybrid Memory Cube and MCDRAM, ran into problems and never became mainstream products. With XBM and ZAM, Chipzilla seems to be trying to course-correct its DRAM ambitions while the wider memory market suffers shortages.
Tan made the comments during the TechSurge: Deep Tech VC Podcast; he said demand for CPUs was huge, particularly for agentic AI and inference.
Chipzilla is looking at new CPU architectures to meet customer needs. It is not just thinking about processors, but how CPUs and stacked DRAM could sit closer together.
Tan said many of today’s memory requirements are not being met. That gives Chipzilla a chance to poke around where the old memory business once bored investors silly.
The key line was his admission that he once avoided memory because it looked like a commodity business. AI has changed that calculation.
Tan said Chipzilla had something in the works but was not ready to “unfold” it. That sounds like corporate teasing, but at least it beats another roadmap slide. He said he thinks about short-term, mid-term and long-term requirements, which suggests this is not just a quick punt on the latest shortage.
Current timelines put Chipzilla’s ZAM project around 2029 to 2030. XBM is expected near the start of the next decade.
Reports suggest memory shortages could persist beyond 2030 as compute and memory demand grows with gigawatt-scale AI factories coming online. If Chipzilla really returns to memory, its foundry services business could get a serious boost.
It would let the company pitch advanced semiconductors, memory solutions and packaging technology under one roof, which is exactly the sort of vertical stack AI customers are throwing money at.
