Crucial DDR5 memory modules, photographed on 22 June 2022. File photo; not CXMT’s proposed server module or HBM. Photo: Rainer Knäpper (Smial) / Wikimedia Commons / Free Art License 1.3. Resized; responsive display may crop. The photograph remains under the same license.
Samsung Electronics closed down 6.79% on October 13, while SK hynix fell 9.47%. The selloff in South Korea’s two memory giants sharpened a question for the AI supply chain: even if demand keeps growing, how much profit can suppliers retain?
Reports about Chinese rival ChangXin Memory Technologies, or CXMT, added to concerns about future competition. Share prices can adjust immediately to that prospect. Factories and customers take longer to establish whether a new product can deliver.
What has CXMT actually promised?
Yicai’s reporting from an October 10 industry conference attributed a year-end target to CXMT President Cao Kanyu: a server DDR5 RDIMM using a new architecture. An RDIMM is a memory module used in servers. The target is a company claim, not independently verified evidence of large-scale shipments.
Memory provides temporary working space for computers. Fitting more capacity into a given chip area could improve a supplier’s economics, provided manufacturing remains reliable.
Yet a demonstration, customer samples and dependable volume deliveries are different milestones. AHR has not obtained enough evidence to independently establish the new product’s manufacturing yield, customer qualification or shipment scale.
CXMT’s September 20 announcement said its fifth-generation DRAM platform and associated low-power products had entered mass production. That earlier announcement concerns different products; it does not establish that the proposed server module is already in volume production.
DDR5 progress does not settle the HBM race
AI systems use several kinds of memory. DDR5 can serve as system memory. High-bandwidth memory, or HBM, stacks memory chips and uses many connections to move large volumes of data quickly, commonly alongside AI accelerators.
The products have different roles and integration requirements. Progress in DDR5 therefore does not demonstrate equivalent HBM supply capability, or imply an immediate price decline across every AI-memory category.
The relevant question for established suppliers is which products face credible new competition, and how quickly customers can adopt alternatives.
Buyers may see an opportunity
A reliable new competitor could put pressure on suppliers’ prices or require additional spending to protect orders. For cloud operators and Asian server makers, however, another qualified source could improve bargaining power and reduce dependence on individual vendors.
Those benefits depend on consistent deliveries, successful testing and applicable procurement and trade rules. A low price alone is insufficient for infrastructure expected to run continuously.
Nor would lower component costs automatically bring cheaper computers, phones or AI services. Contracts, product mix and manufacturers’ pricing decisions would determine how much reaches customers.
What would change the assessment?
CXMT cannot explain the entire October 13 selloff. Energy costs, interest rates and uncertainty about returns on AI spending also formed part of the market backdrop.
Strong demand could absorb additional supply and preserve suppliers’ pricing power. If supply expands faster than demand, buyers may gain greater leverage.
The next evidence should come from product and shipment disclosures, completed customer qualification, and established suppliers’ selling prices and gross margins. Gross margin measures the share of revenue left after direct costs.
The market has marked down expectations. Orders, deliveries and earnings will show whether the underlying business has changed as sharply.
Sources
Aju Press — October 13 closing market figures
Reuters — October 13 market context (intraday)
Yicai — October 10 conference reporting and company target
CXMT — September 20 product announcement