1月26日，市场研究机构Counterpoint Research在最新研报中预测，非GPU服务器AI芯片——AI ASIC阵 营在近期将经历高速增长，到2027年出货数量将达到2024年的三倍，2028年则有望以1500余万颗的规模 反超GPU。 报告显示，这一爆炸性增长的背后，是对谷歌TPU基础设施的强劲需求、AWS Trainium集群的持续扩 展，以及Meta(MTIA)和微软(Maia)随着其内部芯片产品组合的扩展而带来的产能提升。 就人工智能超大规模数据中心的出货量而言，谷歌预计将在2027年之前保持市场领先地位，这主要得益 于其Gemini生态系统的爆炸式增长。 针对谷歌的统治地位，Counterpoint Research研究员David Wu强调："尽管由于市场规模不断扩大，以及 竞争对手(如博通、Marvell和Alchip等设计公司)纷纷采用自研芯片，预计谷歌的市场份额将在2027年下 降至52%，但其TPU集群仍将是无可争议的行业核心和标杆。这一基准的支撑源于训练和运行下一代 Gemini模型所需的庞大且持续的计算能力，而这又需要谷歌持续、积极地扩充其内部芯片基础设施。" 该机构认为， ...

Core Insights - The article discusses the competitive landscape for NVIDIA, particularly focusing on the recent approval by the U.S. government for NVIDIA to sell high-end H200 GPU chips to China and other approved clients, albeit with a 25% sales commission [4][5]. - Despite this approval, NVIDIA faces significant competition from major hyperscalers like Google, Amazon, and Microsoft, who are accelerating their development of self-designed AI chips [5][6]. Group 1: NVIDIA's Market Position - NVIDIA's market share in the AI GPU sector has drastically declined from 95% to nearly zero in the Chinese market due to previous export restrictions [4]. - The company's data center revenue reached $130 billion in the most recent fiscal year, but it is heavily reliant on a few major clients, with the top two clients accounting for 39% of revenue [5][6]. Group 2: Competitors' Developments - Amazon's new AI chip, Trainium 3, is designed to be a low-cost alternative to NVIDIA's GPUs, boasting training speeds four times faster than its predecessor and reducing costs by 50% [8][9]. - Google has released its seventh-generation TPU, Ironwood, which offers a tenfold performance increase over its predecessor and is optimized for high throughput and low latency [11][12]. Group 3: Market Dynamics - The article highlights a shift in the AI chip market, with major companies moving towards self-designed chips, which could potentially capture up to 25% of the market share from NVIDIA [22]. - Amazon aims to increase its self-designed chip usage to 50% and expand its AI cloud market share from 31% to 35% [20]. Group 4: Future Outlook - The competition between performance and cost is expected to intensify by 2026, as NVIDIA maintains a performance edge while competitors emphasize cost savings [17][19]. - The article suggests that while NVIDIA currently dominates the market, the increasing adoption of self-designed chips by major players could significantly alter the competitive landscape in the coming years [22].