Some Thoughts on Why a Powerful Google Does Not Mean Disrupting NVIDIA

• According to Google’s Strongest AI Chip Ironwood Is Coming: Challenging NVIDIA: Google Ironwood TPU single-chip peak computing power reaches 4614 TFLOPS (FP8 precision), and the computing cost is only 1/5 of NVIDIA’s GPU.
• According to NVIDIA CEO Predicts Global AI Data Center Market Size Will Exceed $200 Billion by 2025: NVIDIA occupies 92% of the data center GPU market share and has locked in $500 billion worth of AI chip orders for 2025-2026.
• Research shows (Source: Google TPU VS NVIDIA GPU: Comprehensive Technical Comparison): TPU uses a dedicated ASIC architecture with high energy efficiency but a closed ecosystem; NVIDIA GPU’s general-purpose architecture and CUDA ecosystem form strong barriers.

• Reddit user: AI computing power demand has exploded due to Jevons’ paradox; optimistic about supply chain opportunities such as optical modules (e.g., LITE) and NAND flash memory, and believes AI is not static software, refuting the bubble theory.
• Xueqiu user (Source: Xueqiu Post): Although Google’s TPU is strong, it cannot disrupt NVIDIA because it lacks the CUDA ecosystem and general computing capabilities; recommends focusing on optical modules (Xuchuang, Shenghong) and energy infrastructure.

Research and social media views are consistent: Google’s TPU has advantages in specific scenarios, but NVIDIA’s ecosystem and general-purpose architecture make it difficult to replace. The exponential growth of AI computing power demand will continue to benefit supply chain segments (optical modules, NAND). In terms of investment, NVIDIA remains a core asset, and one can also consider investing in enterprises related to Google’s TPU supply chain.