Samsung's packaging lines are full. So DDR5 and SSD assembly is moving outside.

Samsung's packaging lines at its Cheonan and Onyang plants in South Korea are physically full of HBM work, the vertically stacked memory that sits next to the GPU inside an AI accelerator. To free up space, Samsung is rerouting every new DDR5 module and SSD assembly to outsourced partners, according to industry sources cited by TechRadar and originally reported by TheElec.

Samsung's wafer fabs are unchanged. What is moving is the back-end step: soldering finished memory dies onto circuit boards, sealing them into DDR5 modules and SSD casings, and running the final tests. That work, which the source describes as simpler than advanced HBM packaging, is the part Samsung is handing off.

The three partners are already scaling to absorb it. Dreamtech began mass-producing DDR5 modules at its Indian plant in November and is targeting eventual annual output of more than 50 million units from its Noida facility. Hanyang Digitech has invested more than 31.5 billion won so far in 2026, roughly $22 million at recent exchange rates, to automate its Vietnam memory module plant. SFA Semicon is relocating DDR5 testing and assembly equipment from Samsung's Onyang campus to a new site in the Philippines, with the transition starting in November.

Back-end packaging for HBM, which stacks eight, twelve, or sixteen DRAM dies vertically and bonds them to a base logic die, requires more machines, more cleanroom space, and longer test cycles than sealing a single DDR5 module into a stick. The Cheonan and Onyang lines do both, but the HBM queue is now longer than the DDR5 queue. Samsung's choice, in the source's framing, is to retool the existing footprint for HBM and let outside partners handle the simpler assembly.

The economics make the trade straightforward. DDR5 module assembly is a high-volume, low-margin process that partners in lower-cost geographies can run at competitive cost. HBM packaging, by contrast, has become the scarce input for AI accelerators and commands a structurally higher price. Samsung does not need to publish a per-die margin to make the calculation visible: the HBM side of the Cheonan and Onyang lines is the bottleneck, and the DDR5 side is what can be moved.

It is the second time a major memory maker has made the same trade. Micron has already moved the same kind of conventional memory packaging work to outsourced partners, which is the reason TechRadar frames Samsung's move as following Micron's lead. The structural signal: when two of the three DRAM giants independently reroute back-end packaging from commodity memory to HBM, the constraint is at the packaging step, not at the wafer fab.

Memory makers have always moved assembly to outsourced partners when margins on commodity parts compressed, and the same playbook would apply whether or not AI existed. That reading does not survive the packaging complexity comparison. HBM stack assembly is not a job DDR5 module lines can do, and DDR5 module assembly is not a job HBM lines need to do, so reallocating the simpler work to outsourced partners is the only available lever when the complex work is oversubscribed.

That framing is what makes the SanDisk question worth asking. SanDisk, spun out of Western Digital, sits in a similar position: a NAND- and SSD-focused memory company facing the same AI-driven pull on its back-end capacity. Whether SanDisk follows the same outsourcing pattern is not addressed in the source material and would require its own reporting, including how its NAND-heavy product mix and back-end footprint compare to Samsung's.

Two things to watch. First, the partner capacity numbers (50 million DDR5 units a year at Dreamtech's Noida plant, roughly $22 million of Vietnam automation at Hanyang Digitech, a Philippines equipment move at SFA Semicon) are public, but the actual ramp timeline is not. Second, Samsung's Q3 2025 results break out the HBM and conventional memory segments separately, and any change in the mix between Q3 and the next quarter will show whether the Cheonan and Onyang retool is actually shifting revenue toward HBM or just freeing space that takes time to fill.

The Micron precedent suggests this is the new baseline for memory packaging, not a one-company anomaly. The question is how fast the rest of the industry catches up.