The US is moving to bar federal agencies from buying certain semiconductors tied to major China-based chipmakers, widening procurement restrictions even as memory shortages and rising prices strain electronics supply chains.
FocalTech Systems, a display driver IC (DDI) supplier, held an earnings conference on February 26. Chairman Genda Hu said 2025 operations came in below expectations, as the fading effect of China's subsidy programs and memory shortages weakened smartphone demand, with possible headwinds continuing into 2026. The first quarter is a seasonal low point, with demand expected to recover gradually from the second quarter.
Broadcom has begun shipping the industry's first 2nm custom compute SoC built on its 3.5D eXtreme Dimension System in Package (XDSiP) platform to Japan's Fujitsu, marking a concrete step from roadmap promise to commercial deployment in the AI infrastructure race.
Meta's push to design its own AI chips has reportedly hit major technical and strategic setbacks, forcing the company to scrap its most ambitious in-house training processor and lean more heavily on external suppliers, according to The Information.
A trilateral semiconductor model is emerging, combining Japan's capital, Taiwan's ecosystem expertise, and India's talent. Alongside this, companies including Foxconn, Polymatech Electronics, Nvidia, AMD, Kaynes Semicon, and IBM are deepening India investments, reflecting rising localization, supply-chain ambitions, and expanding AI, packaging, and materials ecosystems despite policy and trade uncertainties.
Taiwan's IC design landscape is undergoing a massive structural shift. Early 2026 revenue data reveals a dual-track performance: while established consumer giants navigate a high-base stabilization phase, specialized leaders in Intellectual Property (IP) and AI-optimized storage are capturing explosive value from the ongoing AI infrastructure wave.
AI chip startup SambaNova Systems has introduced its fifth-generation processor, the SN50, positioning it as a direct alternative to Nvidia’s Blackwell B200 for large-scale AI inference. The company claims up to 5x peak speed in agent-based workloads and up to an 8x total cost advantage in certain deployments.
The boom in cloud-based artificial intelligence (AI) is reverberating far beyond the most advanced chipmaking nodes.
