Asia's Memory Makers Shift Focus as DRAM Margins Challenge HBM Dominance

Memory Market Dynamics Reverse Leadership

Samsung Electronics and Micron Technology are capturing disproportionate gains from surging conventional DRAM and NAND flash prices, outpacing SK Hynix despite the latter's leadership in high-bandwidth memory. SK Hynix saw its DRAM market share contract from 29 percent to 26 percent in the second quarter of 2026, while Samsung expanded to 39 percent and Micron reached 25 percent, according to data from Counterpoint Research.

The shift reflects a fundamental profitability challenge emerging in the HBM segment. DDR5 conventional DRAM now commands approximately $20 per gigabyte, compared to roughly $12 to $16 per gigabyte for HBM3E. Because HBM production consumes significantly more wafer capacity per unit, the margin advantage that once made high-bandwidth memory attractive to manufacturers has begun to erode.

Samsung reported average selling price increases of 44 to 46 percent for DRAM and 67 to 69 percent for NAND in the second quarter. Micron posted even steeper gains, with DRAM ASPs rising over 60 percent and NAND climbing more than 80 percent. SK Hynix recorded more modest increases of approximately 30 percent for DRAM and 55 percent for NAND.

Capacity Constraints Tighten Into 2027

Memory buyers are already facing allocation pressures for 2027, with DRAM and HBM capacity reportedly fully committed. Customers are receiving only 60 to 70 percent of requested volumes, often required to place deposits months in advance to secure supply.

Adata chairman Simon Chen estimates that HBM and AI server applications could consume nearly 70 percent of total DRAM production capacity. SK Group chairman Chey Tae-won projects AI chip demand will increase 60 to 100 percent in 2027, further tightening supply across memory segments.

Long-term supply agreements spanning three to five years with more than ten major customers are now standard practice in the industry. These contracts may moderate average selling price growth from the second half of 2026 through 2027, as fixed pricing terms lock in rates negotiated during the current allocation crunch.

NAND flash remains less constrained than DRAM, though major suppliers are accelerating efforts to secure next year's capacity as AI infrastructure builds continue across data center operators in Singapore, Tokyo, and Seoul.

Packaging and Lithography Advances Shape Roadmaps

Intel announced that yields for its EMIB-T advanced packaging technology have reached 90 percent, though substrate yields remain near 50 percent. The company is targeting mass production in 2027, with projected costs 40 to 50 percent below TSMC's CoWoS packaging. Intel is accelerating construction of its Ohio fabrication complex, offering overtime bonuses to keep the $28 billion project on schedule for 2031 completion.

TSMC is developing an EMIB-equivalent approach as AI demand strains existing CoWoS capacity, reflecting competitive pressure in advanced packaging across Asia's semiconductor ecosystem.

MediaTek confirmed its second-generation data center ASIC remains on track for 2028 mass production, with Intel's EMIB packaging already achieving strong yield levels. The company's first-generation chip is scheduled to ramp in the fourth quarter of 2026, targeting $2 billion in revenue. MediaTek raised its 2027 AI accelerator serviceable addressable market estimate to $80 billion and set a market share target of 15 to 20 percent.

On the lithography front, Inversion Semiconductor is developing a particle accelerator platform intended to surpass ASML's extreme ultraviolet capabilities. The US Department of Energy awarded Inversion $750,000 for an AI digital twin project with Lawrence Berkeley National Laboratory, supplementing $500,000 from Y Combinator and Entrepreneur First. The company aims to deliver a photomask metrology product within two years, though commercialization would require competing against ASML's $150 million to $400 million tools and established relationships with TSMC, Intel, Samsung Electronics, and Zeiss.

Trade Restrictions Extend to Optical Transceivers

The Trump administration is considering restrictions on new Chinese optical transceiver models, extending existing AI export controls into data center networking infrastructure. Zhongji Innolight, which holds approximately 27 percent of the global data center optical module market and supplies Nvidia and Google, faces the greatest potential impact.

Such measures could create opportunities for Lumentum, Coherent, Applied Optoelectronics, Nokia, Ciena, and Cisco. However, the policy risks tighter supply, higher costs, and slower qualification cycles as AI networks transition from 800 gigabit to 1.6 terabit connectivity standards.

The proposal reflects broader US efforts to restrict China's access to critical AI infrastructure components, following earlier controls on advanced semiconductors and manufacturing equipment.

Solar Supply Chain Expansion Targets Polysilicon

Elon Musk is building domestic solar manufacturing capacity through Tesla and SpaceX, targeting up to 200 gigawatts across wafers, cells, and modules, with polysilicon production under consideration. Tesla is planning a 100 gigawatt hub in Texas, while SpaceX is developing 10 gigawatts of heterojunction cell capacity and 50 gigawatts of wafer production.

Chinese suppliers including JSG, Maxwell, and Linton Technologies are involved in the buildout. Musk is exploring non-China polysilicon sourcing and evaluating up to 400,000 metric tons of domestic capacity, reflecting supply chain diversification efforts across the renewable energy sector.

The initiative underscores growing interest among US technology companies in securing vertically integrated supply chains for critical materials, mirroring strategies already established in semiconductor and battery production.