On March 11 (US time), Nvidia announced a US$26 billion five-year investment in open-source large language models and introduced Nemotron 3 Super, its most powerful hybrid mixture-of-experts model to date. The company said the model outperforms OpenAI's open-source GPT-OSS in several benchmark tests.
Geopolitical tariff conflicts combined with the outbreak of Middle East hostilities have pushed even the booming AI server and data center industry—and its supply chain—into a subtle deadlock. Supply chain insiders reveal that despite ample funding, major cloud service providers (CSPs) are hesitating to back suppliers' capacity expansions amid shortages.
Chinese server maker Dawning Information Industry Co. (Sugon) has introduced scaleFabric, its first fully self-developed 400G native remote direct memory access (RDMA) data center networking architecture, aimed at hyperscale AI clusters. The platform integrates 112G SerDes IP, switching chips, network interface cards and management software, all developed domestically, reflecting the company's push to build an independent high-speed AI interconnect stack.
As AI and advanced semiconductor manufacturing drive unprecedented demand for computing power, the reliability of the electricity supply has become a critical concern for both data centers and chip fabrication plants.
Buoyed by sustained high capital expenditure by global cloud service providers (CSPs), AI servers and data center infrastructure have become the ICT supply chain's primary growth drivers over the past two years. Demand for high-performance computing platforms, rack-level integration, and power and thermal architecture in data centers continues to rise, alongside growing demand for ASIC platforms.
Aggressive AI investments by major US cloud service providers (CSPs) are driving a surge in demand for Taiwan's semiconductor and ICT hardware, pushing exports, prices, and private investments toward record levels despite supply constraints.
Meta recently announced plans to launch four generations of ASIC product lines over the next two years, including the already mass-produced MTIA 300 and upcoming MTIA 400, 450, and 500 chips. These ASICs will primarily target AI inference workloads, while AI training tasks will continue relying on Nvidia solutions.
Driven by US President Donald Trump's push for American reindustrialization and efforts to shift industries away from China through high tariffs, Taiwan's economy is set to outpace China's growth for the second consecutive year in 2026.
Wistron Group posted record revenue and earnings per share in 2025, driven by strong demand for AI servers. Still, its gross margin fell as higher unit prices and material costs tied to increased shipments of AI server racks weighed on profitability.
Mech-Mind Robotics, a leading provider of 3D vision systems and software for industrial robots, laid out its plans at Automation World (AW) 2026 to become a global leader in embodied AI. It provides innovative solutions that empower robots with "eyes" and "brains." Through combining advanced 3D cameras with powerful software for perception, path planning, and deep learning, Mech-Mind enables flexible automation and unlocks the full potential of robotics across diverse industries globally.
With artificial intelligence (AI) demand surging and key components such as memory in tight supply, Micro-Star International (MSI) chairman Joseph Hsu and president Jeans Huang said the company is actively expanding in the AI server market, achieving growth of 50 to 100% in 2025, and expects to maintain strong growth in 2026. In addition, they expect gaming products to see a price increase of about 15 to 30% in 2026 due to the impact of rising memory costs.
More coverage
