Overview

This Foundry Monthly Intelligence Report provides timely, always-on market intelligence to keep you aligned with the latest developments in the foundry ecosystem. It includes ad hoc deep dives into emerging trends and market shifts, along with concise analyst takeaways that deliver real-time, decision-ready insights to support strategic planning and execution.

Table of Contents

Samsung’s Turnkey Foundry Model Gains Relevance as Anthropic Diversifies AI Silicon Supply

• Our channel checks show that Anthropic and Samsung are discussing the potential use of Samsung Foundry’s SF4 process for a next-generation AI silicon program. It reflects a broader effort by AI companies to reduce their dependence on TSMC and secure additional manufacturing capacity.
• Samsung’s competitive proposition extends beyond wafer fabrication. Its vertically integrated capabilities across foundry, HBM, and advanced packaging allow it to offer customers a bundled manufacturing solution, potentially improving supply-chain coordination and reducing dependence on multiple external partners.
• Anthropic’s evolving heterogeneous compute strategy could create a clearer entry point for custom silicon. Rather than assigning all workloads to a single architecture, Anthropic appears increasingly willing to allocate pre-training, post-training, and inference across different platforms. This could allow a Samsung-manufactured accelerator to target a defined workload instead of immediately replacing established GPU and TPU infrastructure.

Silicon Capacitors Become a Strategic Layer as Taiwan Foundries Build Dedicated Capacity 

• Silicon capacitors, offering roughly five times the capacitance density of MLCCs with far lower parasitic inductance, are moving into the packaging of AI accelerators.
• TSMC pioneered technology in smartphone processors, but the focus has shifted. PSMC, Winbond, and UMC are piling up dedicated capacity in partnership with chip vendors and passive component makers.
• We expect Taiwan foundry silicon capacitor capacity to scale from ~10 KWPM in 2026 toward 25-35 KWPM (excl. TSMC) in total by the end of 2027, with Google TPU, AWS Trainium, and NVIDIA platforms as the anchor products.

TSMC's EMIB-like 2.5D Packaging: Defending the CoWoS Franchise as Bridge Architectures Gain Ground 

• TSMC's CoWoS supply-demand gap remains ~20% in 2026 and should narrow to roughly 10% through 2027 as in-house capacity scales toward 180 KWPM.
• Intel's EMIB-T design wins prove its success: Google's next-generation TPU and MediaTek's AI ASIC programs have both adopted bridge-based packaging, with further AI/HPC customers in active evaluation.
• We believe TSMC's internally designated EMIB-like program, developed with Taiwanese substrate maker Kinsus, and reflects a structural view that no single supplier can fulfill 2026-2027 demand.

2026 China Local AI GPU Shipments Forecast & Wafer Supply/Demand Analysis

• Our estimation shows that shipments of China's local AI GPUs might reach ~1.5 million units in 2026, while Huawei Ascend, Hygon, and Cambricon hold the top 3 positions with a wafer share of over 95% in total.
• The booming local demand for local AI chips drives a tight supply situation of SMIC N+2 process, which is expected to last throughout 2026 and is likely to be relieved somewhat in 2027.

If you are interested in subscribing to the full report or would like to learn more, please contact us at [email protected].