"As demand for AI servers rises, the adoption of HBM4 memory is set to increase costs, with estimated price premiums exceeding 30%. HBM4 features significant changes, including doubling IO channels to 2048 and using logic chip architecture, leading to higher data transfer rates and reduced latency. However, these advancements also raise manufacturing costs, with wafer costs reaching $7000-8000, making it 3-4 times more expensive than traditional DRAM. In 2026, total HBM shipments are forecasted to exceed 30 billion Gb, with SK hynix expected to maintain over 50% market share. Samsung and Micron need to improve yield rates and production capacity to compete in the HBM4 market.
"TSMC's North America Technology Symposium unveiled a roadmap for advanced packaging solutions, showcasing the 3DFabric® platform, which includes wafer level 3D stacking and advanced packaging technologies. These innovations address bottlenecks in traditional single-chip designs, driven by HPC and AI's rising demand for computing power. Notably, the SoW technology, expected by 2027, aims to boost AI computing power by 40 times. While challenges such as technology maturity, supply chain stability, and high R&D costs exist, TSMC's advanced packaging roadmap and comprehensive strategy underline its commitment to advancing semiconductor technology in the post-Moore's Law era..
"According to the latest supply chain survey from TrendForce, new US export regulations starting in April 2025 will require licenses for exporting semiconductor chips like NVIDIA H20 and AMD MI308 to China. This requirement will significantly impact NVIDIA's ability to sell their H20 and similar chips to China’s AI market, driving the market towards domestic alternatives and a clearer division between the AI server supply chains in China and the US.
In response to these restrictions, Chinese CSPs and OEMs are expected to increasingly rely on domestic suppliers for AI accelerators. Major Chinese CSPs such as Baidu, ByteDance, Alibaba, and Tencent are anticipated to boost their investments in developing in-house AI chips. As a result of the new export controls, the share of externally purchased AI chips (e.g., from NVIDIA and AMD) in China’s AI server market is projected to drop from 63% in 2024 to 41.5% in 2025.
Huawei and Cambricon lead the domestic AI chip sector and are expected to drive the adoption of local chips in 2025. Additionally, Chinese CSPs like Alibaba and Baidu are actively developing their ASICs to replace NVIDIA GPUs. Overall, geopolitical risks are expected to expedite China's autonomy in the AI supply chain.
"TrendForce's 2024 OSAT report highlights packaging's rising importance due to Moore's Law limits and chip complexity. Asian firms dominate, with Taiwan maintaining leadership via technology and capacity advantages. China OSATs are rapidly rising, fueled by policy support and domestic demand, showing significant growth and challenging the established order. US players hold strategic value. Advanced packaging (AI-driven) and geopolitics are key influencing factors. The market shows stable mature leaders coexisting with strongly emerging regional forces, particularly from China, setting the stage for the next phase of industry competition.
TrendForce reports explosive AI computing demand drives CSPs to accelerate in-house ASIC development for cost control and competitiveness. Global AI server shipments are projected ~25% YoY growth in 2025. US CSPs are taking the lead in AI ASIC development: Google (TPU leader), AWS (fastest growth at 104% YoY), Meta (MTIA), and Microsoft (Maia) are actively advancing next-gen chips. Concurrently, US restrictions have prompted major local Chinese CSPs (e.g. BBAT) to accelerate their R&D on in-house AI ASICs in order to lower dependency on US chips. In-house developed ASICs are now a key CSP strategy focus.
"TrendForce reports the US now requires export licenses for AI chips like NVIDIA's H20 and AMD's MI308 to China, effective immediately. This forces NVIDIA and AMD to recognize potential multi-billion dollar losses. Aimed at preventing supercomputer use, licenses might still be granted for general AI applications. H20 China shipments could fall 20-30% in 2025, impacting the supply chain and risking downward revision for HBM demand (H20/MI308 originally ~5%). The restriction accelerates China's move to domestic AI solutions (e.g., Huawei Ascend), deepening US-China supply chain divergence.
Geopolitical instability has fostered the rising consciousness of sovereign AI in 2025, thus prompting the US, Europe, China, and Japan to successively revitalize their domestic chip industries. These efforts will focus on the development and production of autonomous chips and advanced packaging technologies. This report probes into the technical challenges and opportunities on restructuring the industry chain, as well as the profound global impact on the semiconductor landscape under this trend, whilst analyzing the key strategic positions in the new AI chip competition.
"TrendForce reports US tariff policy shifts are impacting the memory market. A 90-day grace period announced in April spurred a Q2 surge in DRAM and NAND Flash orders as buyers and sellers rushed transactions to mitigate future uncertainty. This is expected to drive larger-than-anticipated Q2 contract price hikes. However, this demand pull-forward may weaken market momentum and subsequent price increases in H2 2025. The ultimate direction of US tariff policies remains the key variable for the H2 memory market supply, demand, and pricing.
"TrendForce reports that the introduction of new US tariffs, particularly the 20% ""fentanyl tariff"" on China-made goods, has significantly disrupted the smartphone industry, pushing brands to accelerate supply chain diversification away from China. Apple, which holds a dominant market share in the US and still sources a large portion of its US-bound phones from China, is expected to face the greatest challenges, despite its ongoing efforts to expand production in India.
In contrast, Samsung is better positioned during the 90-day grace period, benefiting from its pre-existing diversified production network in Vietnam and India. However, Lenovo (Motorola), with its heavy reliance on China, faces potential supply bottlenecks and heightened vulnerabilities.
While China remains the core global production hub—accounting for the majority of smartphone manufacturing—the shift toward diversification has become unavoidable. Nevertheless, this transition is not without its challenges, as it introduces higher costs and increased management complexity to global supply chains.
TrendForce reports Mexico remains a key location for server and AI production, particularly for US-bound orders, leveraging USMCA exemptions despite US tariff volatility. The new policy favors USMCA-compliant Mexican exports but restricts China-sourced components. ODMs adopt a ""division of labor"" strategy: Mexico for system assembly, Southeast Asia for modules. While direct tariff impact on AI servers appears manageable (due to US value clause & USMCA), policy uncertainty and economic risks could lead to conservative customer procurement. Mexico's role as a core server supply chain hub is strengthened short-term, but ongoing policy monitoring is crucial.
