Below are the most-read DIGITIMES Asia stories from the week of March 30-April 5, 2026:
US Chip Security Act ends China's special AI chip supply
The proposed US Chip Security Act marks a major escalation in tech controls on China by requiring high-performance AI chips to include hardware-level location verification and tamper-detection systems, enabling regulators to monitor and potentially disable chips if misused. This shifts export controls from performance thresholds to real-time oversight, limiting companies like Nvidia from relying on modified "compliant" chips for the Chinese market while increasing compliance costs and scrutiny.
For China, such embedded monitoring is seen as incompatible with data sovereignty and cybersecurity priorities, especially in sensitive sectors, accelerating a push to eliminate reliance on US-origin chips and build domestic alternatives.
As a result, the global AI supply chain is expected to split into two ecosystems—one led by US-controlled technologies with strict oversight, and another driven by China's efforts toward self-sufficiency—effectively ending the viability of transitional solutions like Nvidia's H20 chips in China.
TSMC plans 12 fabs in Arizona as supply chain shifts from passive to active
TSMC's rapidly expanding US investment—now reaching roughly US$165 billion and potentially encompassing 12 fabs and advanced packaging facilities in Arizona—is driving an unprecedented mobilization of Taiwanese semiconductor suppliers, who are accelerating local hiring, visa applications, and capital commitments to support the buildout.
Backed by US incentives, Taiwan government support, and expectations of broader semiconductor demand, companies across cleanroom, engineering, and equipment segments are shifting from cautious observation to active expansion, aiming to replicate the Hsinchu Science Park ecosystem in Arizona within a decade.
However, despite strong policy support and long-term strategic alignment, suppliers face significant challenges, including 2-3x higher costs, regulatory complexity, and low near-term profitability, underscoring that participation is driven more by future positioning in a globalized supply chain than immediate financial returns.
Samsung reportedly aims to begin silicon photonics mass production in 2028
Samsung Electronics is targeting mass production of silicon photonics (SiPh) by 2028 as part of a broader strategy to reshape AI infrastructure and compete more directly with TSMC. The company plans to integrate photonic and electronic chips by 2027, move to mass production in 2028, and by 2029 incorporate SiPh into advanced packages combining AI processors, GPUs, and high-bandwidth memory, creating a fully integrated platform spanning foundry, packaging, and memory.
This approach aims to reduce costs and production time while attracting major global clients, positioning Samsung to close its estimated 3-year technology gap with TSMC. As industry players like Jensen Huang and Nvidia accelerate investments in SiPh, Samsung's success in achieving early mass production could significantly influence next-generation data center architectures and mark a decisive phase in foundry competition.
TSMC eyes SiPh breakthrough as industry consensus forms
TSMC is moving its SiPh platform, COUPE, from development into mass production in 2026, marking a major step toward integrating optical communication into advanced semiconductor packaging. Built on 3D-stacked SoIC technology and designed to support co-packaged optics (CPO), COUPE addresses growing AI-driven bandwidth and energy challenges by replacing electrical interconnects with optical links while maintaining high performance in a compact footprint.
Backed by industry alignment on a multi-year roadmap and supported by partners like FOCI and MPI, TSMC is collaborating across the supply chain to solve key challenges in testing, fiber integration, and optical assembly. With early adoption already signaled by players such as Nvidia and momentum similar to the earlier rise of CoWoS, COUPE's commercialization is expected to accelerate the adoption of SiPh in data centers and position it as a critical technology for next-generation AI infrastructure.
Global Foundry 2.0 market climbs to record US$320 billion in 2025 revenue
The semiconductor industry is entering the "Foundry 2.0" era, marked by tighter integration across manufacturing, assembly, and testing, with AI demand driving strong growth to a US$320 billion market, according to Counterpoint Research. Leaders like TSMC continue to dominate, though focus is shifting from wafer capacity to system-level integration, with advanced packaging technologies such as CoWoS becoming critical bottlenecks and differentiators.
Meanwhile, competitors including Samsung Electronics and Chinese foundries like SMIC are gaining momentum, while OSAT players benefit from spillover demand as AI-driven packaging capacity tightens. The transition from Foundry 1.0 reflects a broader ecosystem approach—encompassing foundries, IDMs, OSATs, and suppliers—where advanced packaging and system-level optimization are now central to sustaining growth in the AI era.
AI compression won't ease memory crunch, NAND shortage set to persist
AI-driven demand is pushing the global memory market into a prolonged shortage, with NAND flash and server DRAM supply tightening as server memory demand is projected to grow over 40% in 2026 and exceed half of total storage usage. Industry leaders such as Phison Electronics and cloud players like Alibaba Cloud argue that efficiency technologies like Google's TurboQuant compression will not reduce overall demand, as lower costs instead accelerate AI adoption and data generation.
As workloads shift from training to inference, storage—particularly high-capacity enterprise SSDs—has become a key bottleneck, with expanding model sizes and token processing needs dramatically increasing capacity requirements. This dynamic is driving sharp price increases, margin pressures, and a broader "AI storage supercycle," while long semiconductor build cycles mean supply constraints are likely to persist until at least 2027, reinforcing storage's role as a critical strategic resource in AI infrastructure.
Copper price surge drives quarterly lead frame price hikes
Taiwan's lead frame packaging suppliers—including Chang Wah Technology, SDI, and Jih Lin Technology—are implementing a new round of double-digit price hikes in 2026 to offset surging raw material costs, particularly for gold, silver, and copper, while semiconductor demand recovers and orders rebound.
Following initial increases of 10-30% in late 2025, rising input costs and improved downstream pricing power have enabled further adjustments, with some companies adopting mechanisms to link prices to commodity benchmarks like London Metal Exchange copper rates.
As inventory cycles normalize and AI-related demand strengthens, suppliers are benefiting from improved margins and are expected to sustain steady growth through a combination of higher prices and increased shipment volumes, marking the first significant pricing upcycle for the sector in 4 years.
Article edited by Jack Wu
