Author: ZDNet Korea
Compiled by Deep潮 TechFlow
DeepChaohao Summary: Samsung, SK Hynix, and Micron have collectively abandoned in-house development of CXL controllers, opting instead to purchase fabless chip solutions. The reason is pragmatic: developing expensive integrated modules would cannibalize traditional DRAM orders, and customers prefer lower-cost disaggregated solutions. This is not a retreat in technology, but a calculated business decision—design is outsourced to specialized firms, while manufacturing remains in-house, allowing each party to profit in its own domain.
The world's three largest memory manufacturers have collectively abandoned in-house development of CXL (Compute Express Link) controllers, fearing that aggressively promoting their own chips could undermine their core revenue source—demand for general-purpose DRAM (DIMMs)—and fall into a self-cannibalization trap. These memory giants are rapidly shifting to an outsourced strategy, adopting chip designs from specialized fabless companies.
On July 20, industry news in the semiconductor sector indicated that Samsung Electronics, SK Hynix, and Micron have scaled back or canceled their commercialization plans for CXL expansion device controllers. The vacant space has been filled by fabless companies such as Montage, Astera Labs, and Primemas.
Micron and SK "concede," Samsung "waits": Three companies, three exit strategies
Micron, the U.S. company, was the first to abandon in-house development. Micron shut down its independent controller development line and adopted Primemas’s solution. Primemas’s solutions now appear in Micron’s product catalog.
SK Hynix has recently formally communicated to its key partners its decision to discontinue its in-house development of CXL controllers. The company is reallocating relevant personnel to the next-generation computing memory semiconductor PIM (Processing In Memory) field, concentrating its limited R&D resources on more certain future business opportunities.
Samsung Electronics' proprietary CXL controller is for internal research use only by the development team, which explores unverified areas such as LPDDR (low-power DRAM)-based CXL solutions. For commercial sale, the company purchases controllers from fabless suppliers.
This differs from Samsung Electronics' original plan. Samsung initially intended to launch a module integrating its proprietary controller and CMM, but has since removed the official productization plan for its internal controller from its roadmap.
A semiconductor industry insider familiar with the matter revealed: "Samsung Electronics' internal product planning division has removed the internal controller from the list of formal commercialization projects, retaining it solely as a pure pre-research initiative. Currently, the development team is only exploring short-term commercialization possibilities in areas such as running mobile DRAM (LPDDR) on CXL—areas with uncertain market readiness—effectively shifting to a market exploration strategy."
"Wanting to promote CXL but fearing it will undermine DRAM": The self-cannibalization dilemma
Semiconductor industry news indicates that the three major memory manufacturers originally planned to sell "integrated CXL modules," similar to Samsung Electronics, by combining their proprietary controller chips with DRAM on a single board as high-value finished products.
However, the needs of major data center customers differ. To reduce the substantial costs of infrastructure construction, they seek a "disaggregated" architecture—installing a CXL controller separately on the system motherboard and using readily available, low-cost generic DRAM (DIMMs) in rear slots for reuse.
Industry analysis suggests that conflicts at this business model stage are the primary reason for strategic adjustments. If memory manufacturers invest heavily in developing independent controllers and then aggressively push expensive integrated products, cost-conscious customers will tighten their purse strings. Compounding the issue, demand for mainstream DRAM—which previously generated substantial sales—will also decline. Forcing CXL products could inadvertently undermine the company’s biggest revenue generator, the DIMM market, leading to a self-cannibalization dilemma.
Analysts in the semiconductor industry comment: "For memory manufacturers, it is difficult to aggressively pursue the development of proprietary CXL expansion devices if these end products ultimately compete directly in the market with their core cash-generating DIMM products. This is a prudent decision by management—commercializing internal controllers carries no real benefit and risks conflicting with existing core businesses."
Experts emphasize that this move by the three memory manufacturers should not be interpreted as a retreat from or abandonment of CXL technology and its market. Rather, it is more akin to mitigating risks during the market’s early, uncertain phase and returning to the semiconductor ecosystem’s inherent, efficient division of labor.
Another semiconductor industry professional said: "Rather than unreasonably monopolizing CXL leadership, it is more practical to adopt a division of labor with Fabless companies. As the global semiconductor market becomes increasingly specialized, the optimal ecosystem division—where design is handled by specialized Fabless firms and manufacturing by memory manufacturers—will accelerate further."