Benzinga·2025-12-01 15:43Core Insights - The AI chip competition is escalating, with Alphabet Inc. challenging Nvidia's dominance through its advanced AI model and custom-built chips [1][3] - Alphabet's stock performance has significantly outpaced Nvidia, with a 30 percentage point lead in November and a year-to-date increase of 70% compared to Nvidia's gains [2] Alphabet Inc. Developments - The launch of Gemini 3, Google's latest AI model, is a key driver of its stock rally, showcasing advancements in AI capabilities [3][5] - Gemini 3 is gaining traction with 650 million monthly active users, powered by Google's TPUv7 chips, which are designed to compete directly with Nvidia's offerings [6] TPU vs. GPU Comparison - Google's TPUs are fundamentally different from Nvidia's GPUs, focusing on specific use cases that enhance efficiency and cost-effectiveness [7][8] - TPUv7 outperforms Nvidia's Blackwell Ultra in performance per watt, achieving 5.42 teraflops per watt compared to 3.57 teraflops per watt for Nvidia [9] Cost Efficiency - The rental cost for TPUv7 is estimated at $3.50 per hour, significantly lower than Nvidia's $6.30 per GPU per hour, potentially leading to a total cost of ownership that is up to 40% lower for TPUv7 [9] Market Dynamics - Despite the advantages of TPUs, their limited availability on Google Cloud Platform makes them less appealing compared to Nvidia's GPUs, which are accessible across multiple cloud services [10][11] - Google is projected to spend $10 billion in 2025 on GPUs and TPUs, indicating a strategy for hybrid computing [12] Nvidia's Valuation - Nvidia shares are currently trading at approximately 25 times forward PE, presenting a potential buying opportunity as historical trends suggest rebounds to higher valuations [13] - Nvidia is experiencing a significant discount relative to Broadcom, which may reflect market expectations of a shift in AI market share [14] Conclusion - While TPUs excel in specific performance metrics and cost, Nvidia maintains a competitive edge through its ecosystem, availability, and broad workload support [15]