AI infrastructure, memory supply, advanced packaging and US-China tech restrictions led this week's industry agenda. Below are the most-read DIGITIMES stories from the week of August 3-10, 2026.
Inversion Semiconductor eyes ASML with accelerator X-ray lithography
Inversion Semiconductor is developing a particle-accelerator lithography platform that it says could surpass ASML's EUV limits, with a system 1,000 times smaller than conventional accelerators, 500 million pulses per second, twice the transistor density, and triple wafer throughput.
The US Department of Energy awarded Inversion US$750,000 for an AI digital twin with Lawrence Berkeley National Laboratory, on top of US$500,000 from Y Combinator and Entrepreneur First. A photomask metrology product is targeted within two years, though commercialisation still means taking on ASML's US$150-400 million tools and ecosystem spanning TSMC, Intel, Samsung Electronics and Zeiss.
Intel speeds Ohio fab build, EMIB-T yield hits 90%
Intel is accelerating its Ohio fab build and EMIB-T packaging push, offering overtime bonuses to keep its US$28 billion-plus complex on track for 2031. EMIB-T yields have reached 90%, though substrate yields remain near 50%; mass production is targeted for 2027, with costs potentially 40-50% below TSMC's CoWoS. TSMC, meanwhile, is developing an "EMIB-like" approach as AI demand strains CoWoS capacity.
DRAM price surge lifts Samsung, Micron over SK Hynix
Samsung and Micron are gaining more from surging conventional DRAM and NAND prices than HBM leader SK Hynix, whose DRAM share fell from 29% to 26% in 2Q26, while Samsung rose to 39% and Micron to 25%, according to Counterpoint Research. Samsung's DRAM ASP jumped 44%-46% and NAND 67%-69%, while Micron's rose over 60% and 80%, respectively; SK Hynix posted smaller gains of about 30% and 55%.
With DDR5 reaching US$20/GB versus roughly US$12–US$16/GB for HBM3E, HBM's heavier wafer use is eroding its profitability edge, while 3–5-year LTAs with more than 10 major customers could temper ASP growth from 2H26 through 2027.
Musk builds US solar supply chain, eyes polysilicon
Elon Musk is building a US solar supply chain through Tesla and SpaceX, targeting up to 200 GW across wafers, cells and modules, with polysilicon potentially next. Tesla reportedly plans a 100 GW Texas hub, while SpaceX is developing 10 GW of HJT cells and 50 GW of wafer capacity. Chinese suppliers including JSG, Maxwell and Linton Technologies are involved, while Musk is also exploring non-China polysilicon sourcing and up to 400,000 metric tons of domestic capacity.
Intel EMIB yields hit target, MediaTek eyes 2028 ASIC ramp
MediaTek says its second-generation data-center ASIC remains on track for 2028 mass production, with Intel's EMIB packaging already at strong yield levels, while its first-generation chip is set to ramp in 4Q26 and generate US$2 billion in revenue. The company raised its 2027 AI-accelerator SAM to US$80 billion and its share target to 15%–20%, while approving a US$5 billion flexible financing plan on top of more than US$7 billion in cash to secure capacity and support key suppliers.
2027 DRAM, HBM capacity reportedly sold out as AI demand surges
2027 DRAM and HBM capacity is reportedly already fully allocated, with buyers receiving only 60%-70% of requested volumes and often paying deposits upfront. Adata Chairman Simon Chen estimates HBM and AI servers could consume nearly 70% of DRAM capacity, while SK Group Chairman Chey Tae-won expects 2027 AI chip demand to rise 60%-100%. NAND remains less tight, though major suppliers are also rapidly locking in next year's capacity.
US weighs China optical transceiver ban as AI supply risks grow
The Trump administration is weighing a ban on new Chinese optical transceiver models, extending US AI restrictions into data-center networking. Zhongji Innolight, which holds about 27% of the global data-center optical module market and supplies customers including Nvidia and Google, faces the greatest exposure.
The move could open opportunities for Lumentum, Coherent, Applied Optoelectronics, Nokia, Ciena and Cisco, but risks tighter supply, higher costs and slower qualification as AI networks shift from 800G to 1.6T.
Article edited by Jack Wu
