智通财经网·2026-01-28 07:23Core Insights - The report from BNP Paribas highlights the launch of Microsoft's second-generation self-developed AI chip, "Maia 200," which is expected to trigger a new wave of investment in the AI computing power industry, particularly benefiting leaders in custom AI ASIC chips like Marvell and Broadcom [1][6] - Analysts predict that the market share of ASICs compared to NVIDIA's AI GPU clusters could significantly increase from the current ratio of 1:9/2:8 to nearly equal [1] - The ongoing AI infrastructure investment wave is projected to reach between $3 trillion and $4 trillion globally by 2030, driven by unprecedented demand for AI computing power [4] Group 1: AI Chip Market Dynamics - The trend of self-developed AI chips by cloud computing giants like Microsoft, Google, and Amazon is reshaping the AI ASIC and GPU landscape, with significant implications for data center interconnects and high-speed cabling [2][3] - Google’s TPU AI chip production is expected to surge, with projections of 5 million and 7 million units in 2027 and 2028, respectively, indicating a potential shift towards external sales of TPU chips [7] - The demand for AI ASICs is anticipated to triple by 2027, surpassing GPU shipments, driven by the expansion of Google’s TPU infrastructure and AWS Trainium clusters [7] Group 2: Infrastructure and Connectivity - The report identifies potential beneficiaries in the data center connectivity space, including Amphenol for high-speed copper cables and Lumentum for optical interconnects, as the demand for AI infrastructure grows [8] - The integration of high-performance networking solutions, such as NVIDIA's InfiniBand and Google's Optical Circuit Switching, is crucial for the efficiency of AI data centers, emphasizing the role of both copper and optical interconnects [9][10] - The Maia 200 AI infrastructure is expected to be deployed in a specific topology designed for AI inference workloads, with large-scale deployment anticipated to accelerate in the second half of 2026 [11]