Spent time with a CXL controller engineer and a packaging engineer in the Bay Area today. We talked a lot about disaggregated memory — a theme that's quietly picking up steam.

A few things stood out:

CXL controller companies are broadening their product lines beyond chips into AICs, pooled memory, and JBOM. Customer interest is real and growing.

Both engineers pointed to SK hynix's weak spot: limited SoC design experience. They believe this contributed to SK hynix's recent decision to stop developing its own CXL solution in-house. The same limitation could complicate their custom HBM ambitions. SK hynix knows this and has been aggressively hiring to close the gap.

The CXL engineer raised a fair question: how profitable can custom memory actually be? His take: conventional DRAM might still own the largest profit pool and the best margins long-term. We spent a while unpacking that.

His company is working with Penguin Solutions and expanding ties with SK hynix and Samsung. He expects the overall business to scale significantly over the next few years.

This space is still early, but the supply chain is moving fast. Worth watching how the memory stack evolves — especially as hyperscalers push for more flexible, disaggregated architectures.