According to Bloomberg, nine major U.S. industry associations—including those representing the automotive, medical device, retail, and telecommunications sectors—on June 4 jointly sent a letter to the Treasury Secretary and the Commerce Secretary, warning that: AI data centers are consuming “disproportionate memory chip production capacity,” driving memory prices to unprecedented spikes while causing the supply of both the manufacturing industry and downstream consumer industries to shrink rapidly. Market analysis estimates that HBM (high-bandwidth memory) has accounted for about 25% of DRAM’s overall wafer production capacity.
The specific requests in the joint letter: three government action directions
Based on Bloomberg’s report on the joint letter, the three directions of action the industry associations are asking the government to take include: working with memory manufacturers and downstream buyers to expand capacity in the U.S. and friendly countries; making use of trade agreement mechanisms or the framework of the CHIPS Act to ensure supply-chain security; and clearly requiring that the coverage of保障 policy scope must include consumer-end and manufacturing-industry markets, rather than being limited to AI-related customers.
The CHIPS Act cited in the joint letter is the U.S. semiconductor subsidy and domestic manufacturing policy passed in 2022. Previously, it has mainly been used to encourage foundries such as TSMC to set up plants in the U.S.; the industry associations’ request is to extend policy momentum from the AI compute side down to traditional manufacturing and the consumer end.
Current U.S. capacity status of three major memory companies: confirmed construction plans and actual progress
Currently, Micron Technology is the only company producing memory wafers within the U.S.; Samsung and SK hynix’s main memory-wafer production capacity remains concentrated in their home base in South Korea. Micron is advancing new plant plans in New York State and Idaho, but Bloomberg notes that it will still take several years for these capacities to come online.
The U.S. Commerce Secretary, Howard Lutnick, has put pressure on the two Korean companies, demanding that SK hynix and Samsung build facilities in the U.S. Samsung is currently building a logic chip plant in Austin, Texas, and SK hynix plans to build a packaging facility in Indiana; neither is a memory wafer plant. Micron and SK hynix both surpassed a trillion-dollar market value last month, and since the beginning of the year, both companies’ stock prices have risen by more than 240%.
Frequently Asked Questions
Why can HBM and standard DRAM not share the same batch of wafer capacity?
HBM (high-bandwidth memory) is a special structure that vertically stacks multiple layers of memory chips, placed closely next to AI computing chips. Its manufacturing processes and technical requirements differ from those of standard DRAM. Samsung, SK hynix, and Micron are reallocating wafer production capacity originally used to produce standard DRAM to fulfill HBM orders from AI chip makers such as Nvidia, because HBM has higher profit margins. This directly compresses the amount of standard DRAM supply available for industries such as automobiles and medical devices.
What specific industries are included in the nine U.S. industry associations?
The source text confirms that the covered industries include automotive, medical devices, retail, and telecommunications, and describes them as “major industries spanning both manufacturing and the consumer end.” The joint letter states that it is not peripheral industries being affected, but rather the core backbone of the U.S. economy.
Even if Samsung and SK hynix build plants in the U.S., why can’t the supply gap be solved in the short term?
Samsung’s Austin, Texas construction project and SK hynix’s plan in Indiana are both for logic chip plants or packaging plants—not memory wafer plants—so they cannot directly increase DRAM supply. Micron’s new plant plans in New York State and Idaho are memory wafer plants, but Bloomberg notes that it will still take several years before these capacities are actually put into operation. During this window, the competitive landscape for memory supply between traditional manufacturers and major AI companies will not change.
