Samsung, SK Hynix, and Micron All Abandon In-House CXL Controller Development
0xBroomberg
The world's three largest memory chipmakers have collectively scrapped in-house CXL controller programs and turned to outside suppliers — the core reason is that pushing their own controllers would cannibalize their most profitable business, commodity DRAM modules. Specialist chip-design firms are stepping in.
What is a CXL controller, and why did memory giants want to build their own?
CXL — Compute Express Link, a new interface standard that lets CPUs and memory talk at high speed — is seen as the next-generation memory-expansion architecture for data centers. The controller is the "brain" of that system.
Samsung, SK Hynix, and Micron all had the same playbook: package the controller with their own DRAM on one board and sell it as a complete solution. This means → each sale would carry a "solution premium" on top of the bare memory chips.
In plain terms = all three wanted to move from "selling components" to "selling the whole box." The controller was the key upgrade piece.
How did each company exit?
Micron moved most decisively: it shut down its CXL controller R&D unit entirely and adopted PrimeMass's controller. The product is already listed in Micron's catalog.
SK Hynix formally notified partners that its in-house controller project is terminated. The R&D team has been reassigned to PIM — processing-in-memory, which embeds compute functions directly inside memory chips. This means → SK Hynix chose to bet its headcount on a next-generation architecture rather than keep fighting over controllers.
Samsung was more cautious: it removed its in-house CXL controller from the commercialization roadmap but kept an exploratory research effort. The team now focuses on areas like combining LPDDR with CXL. The previously planned "in-house controller + CXL memory module" bundled product has effectively been shelved.
Why the collective retreat? How does the cannibalization math work?
Data-center customers prefer a modular architecture — the controller sits independently on the motherboard, while memory stays as standardized, low-cost DIMM modules. That preserves procurement flexibility and holds down costs.
In plain terms = customers would rather assemble their own stack than be locked into one vendor's bundle.
One semiconductor industry insider put it bluntly: if a CXL integrated product has to compete head-on with the company's own DIMM business, the case for commercializing the controller collapses. This reflects a fundamental tension — the more the new business succeeds, the deeper it cuts into the old one — and management will naturally protect the higher-margin pillar.
Who is picking up the controller market?
Astera Labs, Montage Technology (蒙太奇科技), and PrimeMass — three specialist fabless chip-design firms — are filling the gap.
This means → the CXL ecosystem has entered a classic specialization phase: controllers go to fabless design houses, while memory makers keep doing what they do best — manufacturing and DRAM process technology.
In plain terms = each side plays to its strength — design stays with designers, manufacturing stays with manufacturers. No one tries to do everything.
What does this collective pivot signal?
The three giants are not abandoning CXL itself; they are abandoning the "controller-to-memory, all in one" business model. This means → R&D resources will be redirected to advanced DRAM process nodes, HBM (High Bandwidth Memory), and processing-in-memory.
For controller firms like Astera Labs, the memory giants' exit turns competitors into customers — opening up addressable market.
Whether the CXL ecosystem can accelerate commercialization under this division-of-labor structure is the key test of whether this strategic pivot pays off.
Content is for reference only, not financial advice.