1. YMTC Plans to Capture 20% Global NAND Supply Amid U.S. Sanctions

Yangtze Memory Technologies Corp (YMTC) is set to start mass production of cutting-edge NAND products at a new fab in Wuhan in the second half of 2026, with plans to build two additional fabs each producing 100,000 wafers per month. Notably, more than 50% of Phase 3's equipment, materials, and tools have been sourced from Chinese domestic suppliers, defying U.S. trade sanctions that restrict access to Western fabrication equipment. YMTC also unveiled its Xtacking 4.0 memory chip design for 3D NAND, implementing separate dies for logic and memory array, and is reportedly allocating around 50% of Phase 3's capacity to DRAM, including HBM packaging development. Industry observers are closely watching whether Chinese memory underdogs can shift global procurement decisions given a 10-20% price edge over Japanese, Korean, and U.S. rivals.

Source: EE Times (April 26, 2026)

2. TSMC to Spend in 2026, Warns Shortages Will Persist Through 2027

TSMC announced it expects to spend nearly billion in 2026 as it invests heavily in production capacity to meet surging AI demand from customers including Nvidia, AMD, and Apple. CEO C.C. Wei warned on a Q1 2026 earnings call that despite the record spending, shortages of advanced technology wafers and packaging will likely persist through 2027, as building and ramping a new fab takes 3-5 years. TSMC plans to rush construction of three new fabs in Japan, Taiwan, and the U.S. for 3-nm chips, with the Taiwan facility starting production in H1 2027, followed by the U.S. site in late 2027 and Japan in 2028. Wei also brushed off Elon Musk's TeraFab plans, saying the fundamental rules of the foundry business--technology leadership, manufacturing excellence, and customer trust--cannot be bypassed.

Source: EE Times (April 26, 2026)

3. SK hynix Receives 2026 IEEE Corporate Innovation Award for HBM Leadership

SK hynix announced it received the Corporate Innovation Award at the 2026 IEEE Honors Ceremony held in New York on April 24. The award recognizes the company's contribution to the global AI computing ecosystem through stable mass production of all generations of High Bandwidth Memory (HBM). Industry observers credited the achievement to SK Group Chairman Chey Tae-won's strategic direction in securing long-term technological competitiveness and expanding AI infrastructure partnerships with global Big Tech firms in the United States. SK hynix President and CDO Ahn Hyun accepted the award, stating the company will continue collaborating closely with global customers and partners to lead AI innovation.

Source: EE Times (April 26, 2026)

4. TSMC Unveils Roadmap: A14 in 2028, Photonics Solution for AI Data Centers

At its annual technology symposium, TSMC revealed its chip roadmap including two derivatives of its upcoming A14 (1.4-nm GAA) process--A13 and A12--both starting production in 2029. The company also announced N2U, a power-saving variant of its 2-nm process, will enter production in 2028 reducing power consumption by up to 10%. Notably, TSMC said it does not yet need ASML's million high-NA EUV tools for its leading-edge nodes, continuing to leverage multi-patterning knowhow on existing low-NA EUV equipment. On packaging, TSMC outlined plans to roll out COUPE (Compact Universal Photonics Engine) for lower power consumption and latency in AI data centers, expand CoWoS interposer sizes from 5.5 reticle this year to 14 reticle by 2028, and indicated the CoWoS supply crunch is ending as capacity has grown substantially to meet customer demand.

Source: EE Times (April 26, 2026)

5. Chip Sovereignty Shifts from Fabs to Full AI System Stack

Exclusive interviews with AMD CTO Mark Papermaster and Tufts Professor Chris Miller (author of *Chip War*) reveal a fundamental shift in semiconductor strategy: as AI reshapes the industry, strategic advantage is no longer about controlling individual fabs but about dominating the full system stack including AI accelerators, advanced packaging, memory bandwidth, and the software that ties them together. Papermaster noted AI workloads now require extensive computation across a broader system spanning CPUs, GPUs, accelerators, memory, storage, and networking. Miller highlighted that many of the most vulnerable chokepoints lie in materials, chemicals, gases, and subcomponents rather than in manufacturing alone. Both emphasized that no country can realistically replicate the entire semiconductor ecosystem, making diversification across regions and specialization in existing strengths the pragmatic path to resilience.

Source: EE Times (April 26, 2026)

6. GUC Demonstrates 12 Gbps HBM4 IP on TSMC 3nm at North America Symposium

Global Unichip Corp. (GUC) announced the successful demonstration of a 12 Gbps HBM4 IP platform implemented on TSMC's 3nm process at the TSMC 2026 North America Technology Symposium. The platform features GUC's in-house HBM4 Controller and PHY IP integrated with a partner's HBM4 memory using TSMC's CoWoS advanced packaging technology. Compared to the previous HBM3E generation, GUC's HBM4 IP delivers 2.5x bandwidth, 1.5x power efficiency improvement, and 2x area efficiency. CTO Igor Elkanovich stated GUC is the first company to demonstrate 12 Gbps HBM4 IP to customers, offering a complete 2.5D/3D IP solution for modern 3.5D system architectures alongside its UCIe, Glink-3D, and SoIC-X on CoWoS offerings.

Source: EE Times (April 22, 2026)

Outlook

The past week reinforced that AI infrastructure demand continues to be the dominant force reshaping the semiconductor industry. TSMC's spending and expansion of 3-nm capacity reflect the severity of the current chip shortage, while YMTC's progress in NAND--achieved largely through domestic suppliers--underscores that U.S. sanctions alone have not halted China's semiconductor ambitions. The strategic conversation is maturing beyond fab-centric thinking: as AMD and industry analysts make clear, the real battleground is the full system stack--memory bandwidth, advanced packaging, and the software ecosystem that ties it all together. With SK hynix recognized for HBM leadership and GUC rolling out HBM4 IP, memory remains at the epicenter of AI competitiveness. Supply constraints across advanced nodes and packaging are expected to persist through at least 2027, keeping foundry capacity allocation a top priority for fabless AI chipmakers worldwide.