NVIDIA's NVLink Fusion Launches, Qualcomm's CPU Comeback, and Taiwan's AI Supercomputer - Silicon Brief #19

Hi there, In this edition of Silicon Brief, we're diving into the latest developments in AI chips and data centers. NVIDIA is making headlines with the launch of its NVLink Fusion for AI infrastructure, a move that aims to enhance connectivity and performance in AI systems. Meanwhile, Qualcomm is collaborating with NVIDIA to develop custom data center CPUs, integrating advanced technology to boost efficiency and processing power. On the data center front, Foxconn and NVIDIA are teaming up to build an AI supercomputer in Taiwan, a project that underscores the growing demand for AI-driven data processing capabilities. Additionally, TrendForce predicts a 14.3% growth in the data center interconnect market by 2025, highlighting the rapid expansion and evolution of data infrastructure to support AI advancements.

Open Compute Project Foundation is working on new data center power architecture to reduce energy losses significantly by introducing '1 Megawatt racks'. Aalo Atomics has introduced the Aalo Pod, a modular nuclear reactor designed to meet the energy demands of AI data centers, aiming to provide a stable and sustainable power source. Google has hired Patrick Taylor, previously at Microsoft, to lead its advanced energy technology team, focusing on carbon-free energy solutions. At COMPUTEX 2025, MiTAC Computing Technology Corp. introduced new AI data center solutions, focusing on energy efficiency and performance with advanced servers and liquid cooling technologies. Vaire, a U.K. startup, has developed a reversible computing chip that reportedly reduces energy consumption for AI workloads by 50% in initial tests.

Red Hat has introduced the llm-d community, an open-source project to enhance scalable generative AI inference, collaborating with industry leaders to improve performance for large language model inference. Wiwynn introduces its new AI servers and cooling technologies at Computex 2025, in collaboration with Wistron, featuring the NVIDIA GB300 NVL72 system. Fabric8Labs and Wiwynn are set to present advanced cold plates using ECAM technology at Computex 2025, aimed at improving cooling solutions for AI data centers.

Nvidia is preparing to deliver a new AI chip to China, designed to meet US export restrictions by using GDDR7 memory. The company also plans to establish a research and development centre in Shanghai. Recent economic agreements during President Trump's Middle East trip include a $142 billion arms deal with Saudi Arabia and the launch of the HUMAIN initiative for AI development, highlighting the role of AI and semiconductors in defense.

Qualcomm plans to develop custom data center CPUs using Nvidia technology, marking its return to the data center CPU market. NVIDIA announces the creation of Taiwan's first AI supercomputer in collaboration with ASUS to advance AI research. Foxconn and NVIDIA partner with the Taiwan government to build an AI factory with a supercomputer using 10,000 NVIDIA Blackwell GPUs.

NVIDIA announced NVLink Fusion at Computex 2025, a silicon solution for AI infrastructure in collaboration with MediaTek and Qualcomm. Marvell Technology, Inc. has partnered with NVIDIA to integrate NVLink Fusion into custom cloud platform silicon for AI infrastructure. Supermicro is offering over 20 AI systems optimized for the new NVIDIA RTX PRO 6000 Blackwell GPUs.