Core Insights - The demand for AI computing is surging, prompting tech giants to accelerate their development of self-developed cloud ASIC accelerators [1][2] - The self-developed cloud ASIC accelerator market is transitioning from a high-growth phase in 2023 and 2024 to a stable growth phase, with an estimated shipment volume of 4.962 million units in 2023, reflecting a significant year-on-year growth rate decline to 23% [1] - The growth is driven by rising costs of training generative AI models, the popularity of small language models, high prices of NVIDIA GPUs, and U.S. restrictions on Chinese companies accessing advanced AI chips [1] Group 1 - Self-developed cloud ASIC accelerators are becoming a key competitive factor for cloud service providers (CSPs) in the generative AI era [1] - These accelerators can reduce energy consumption, help CSPs control AI infrastructure costs and supply chains, and mitigate reliance on NVIDIA or international political factors [1] - Companies like Google, Amazon, Huawei, Microsoft, and Tesla are actively investing in the self-developed cloud ASIC accelerator market [1] Group 2 - By 2025, the competitive landscape of the self-developed cloud ASIC accelerator market will significantly change, shifting from Google’s dominance to a three-way competition among Google, Amazon, and Huawei [2] - Google is expected to maintain its market leadership with a shipment increase of only 1% to 2.551 million units, while Amazon's shipments will rise to 1.369 million units due to the mass production of Trainium 2 and collaboration with Anthropic [2] - Huawei's shipments are projected to double to 800,000 units, driven by domestic market demand and capacity expansion, ranking it third [2] - Future competition in the AI market will extend beyond AI model innovation to control over computing infrastructure, with companies possessing self-developed ASICs being able to allocate computing resources more flexibly and reduce costs [2]