AI基础设施的军备竞赛正以前所未有的速度升级。谷歌最新发布的下一代张量处理单元（TPU）平台Ironwood，以其惊人的性能飞跃，再次推高 了这场竞赛的门槛。 根据谷歌在Hot Chips 2025大会上披露的信息，其第七代TPU架构Ironwood在核心性能上实现了指数级增长，单颗Ironwood芯片的峰值算力高达 4614 TFLOPs。与谷歌2022年推出的TPU v4相比，Ironwood的单芯片算力提升了超过16倍；即便是与去年发布的TPU v5p相比，也增长了近10 倍。 Ironwood的发布不仅是单个芯片的革新，更是一套完整的、旨在实现极致扩展性的系统级解决方案。谷歌同时公布了围绕该芯片构建的机架、网 络互连和冷却系统，展示了其将尖端算力转化为大规模、高效率生产力的全栈能力。 性能飞跃：单芯片算力提升超16倍 谷歌此次公布的数据清晰地展示了其TPU平台性能的演进路线。具体来看，Ironwood的单芯片峰值算力达到4614 TFLOPs，并配备了192 GB的高 带宽内存（HBM），带宽高达7.4 TB/s。与之对比，2022年发布的TPU v4单芯片算力为275 TFLOPs，配备32 GB ...

Core Viewpoint - The AI server market is dominated by NVIDIA, with the emergence of ASIC servers as a significant competitor, indicating a shift in the industry landscape [1][6]. Group 1: Market Growth and Projections - The global server market is expected to grow at a CAGR of 3% from 2024 to 2026, approaching a size of nearly $400 billion by 2026, with AI servers being the main growth driver [1]. - AI server shipments are projected to maintain double-digit growth, while overall server shipments will see a slight slowdown, with a 4% year-on-year increase in 2024 [1]. - High-end GPU servers, particularly those equipped with 8 or more GPUs, are expected to see over 50% growth in 2025 and a low 20% increase in 2026 [1]. Group 2: NVIDIA's Product Launches - The GB200 server began mass shipments in Q2 2025, with expected shipments of approximately 7,000 units, increasing to 10,000 units in Q3 2025 [3][4]. - The GB300 server is set to enter mass production in Q4 2025, with expected shipments in the thousands [2][3]. - The introduction of the next-generation Rubin chip is anticipated to raise the average selling price (ASP) of high-end AI servers, enhancing market size and supply chain opportunities [1]. Group 3: Competitive Landscape - While NVIDIA leads the market, major cloud service providers (CSPs) like Amazon, Meta, Google, and Microsoft are advancing with their ASIC servers, which offer cost and customization advantages [6][7]. - NVIDIA's GB200 chip boasts a BF16 performance of 2250 TFLOPS, significantly outperforming competitors' offerings in terms of performance [10]. Group 4: Future Market Opportunities - Broadcom predicts that the market for custom XPU and commercial network chips will reach $60-90 billion by FY2027, indicating substantial growth potential in the AI server market [8]. - Marvell anticipates a 53% CAGR growth in its data center market from 2023 to 2028, further supporting the upward trend in AI server demand [8].

Investment Rating - The report maintains a "Buy" rating for several key companies in the semiconductor and AI sectors, including 中芯国际 (SMIC), 海光信息 (Haiguang), and others, indicating strong growth potential in the domestic AI and computing landscape [5][6]. Core Insights - The domestic AI landscape is witnessing significant advancements with the emergence of models like 豆包 (Doubao) and DeepSeek, which are leading the charge in multi-modal and lightweight AI model development, respectively [1][2]. - The report highlights a shift towards domestic computing power solutions, with chip manufacturers rapidly adapting to the evolving AI ecosystem, particularly through advancements in semiconductor processes and AI training capabilities [2][3]. - There is a notable increase in capital expenditure among cloud computing firms, driven by the rising demand for AI computing infrastructure, which is expected to lead to a "volume and price rise" scenario in the cloud computing market [3][4]. Summary by Sections Section 1: Breakthroughs in Domestic AI Models - 豆包 has emerged as a leading multi-modal model, enhancing capabilities in speech, image, and code processing, with a significant release of its visual understanding model in December 2024 [1][11]. - DeepSeek focuses on lightweight model upgrades, achieving a remarkable cost-performance ratio with its DeepSeek-V3 model, which has 671 billion total parameters and costs only 557.6 million USD, positioning it among the world's top models [1][12]. - The rapid iteration of domestic models, including updates from 通义千问 and others, reflects a competitive landscape that is accelerating the development of AI applications [1][34]. Section 2: Advancements in Domestic Computing Power - 中芯国际 is advancing its semiconductor processes, with N+1 and N+2 technologies being developed to support the growing demand for AI chips, achieving significant performance improvements [2][56]. - The report notes that the domestic chip industry is evolving, with companies like 昇腾 (Ascend) and others making strides in AI training and inference capabilities, thereby reducing reliance on international competitors [2][59]. - The cloud computing sector is experiencing a capital expenditure boom, with companies like 华勤 and 浪潮 rapidly deploying servers that are compatible with domestic computing power solutions [3][4]. Section 3: Infrastructure and Supply Chain Developments - The report emphasizes the need for enhanced computing infrastructure to meet the surging demand for AI applications, with significant investments being made in server and power supply innovations [3][4]. - Innovations in power supply and cooling systems, particularly the shift from traditional air cooling to liquid cooling, are becoming essential to support the increasing power density in data centers [4]. - The report identifies key players in the supply chain, including companies in power supply, cooling, and server manufacturing, that are poised to benefit from the growth of the AI and computing sectors [5].