Google vs. NVIDIA: AI Chip Competition and Investment Insights

:

• According to 谷歌 TPU 与英伟达 GPU 在 AI 推理的性能对比: Google’s TPU v7 achieves 4614 TFLOPS per chip and 42.5 Exaflops cluster算力, matching NVIDIA’s GB300 but with better energy efficiency.
• Research shows: NVIDIA holds ~80% AI chip market share due to CUDA ecosystem, with Blackwell architecture in production and Rubin series expected in 2026 美国芯片公司市值十强｜AI芯片竞争白热化.

:

• Reddit users note Google’s TPU+OCS优势 but reliance on NVIDIA for flexibility; Anthropic’s 1M TPU deal validates demand.
• 雪球帖子 argues TPU is closed-stack, lacks CUDA compatibility, so NVIDIA remains core for other companies.

: Google’s technical progress is real but NVIDIA’s CUDA ecosystem is a strong barrier. Short-term NVIDIA leads, long-term Google’s full-stack could challenge. Investment opportunities in supply chains like LITE (Lumentum), 旭创 (300308), 胜宏 (300476), NAND, and energy storage.