Google's Ironwood TPU vs NVIDIA: Competitive Dynamics and Investment Implications

• According to research: Google’s Ironwood TPU single-chip performance reaches 4.6 petaFLOPS FP8, energy efficiency is 2-3 times higher than GPUs, yet it only holds a 3-4% share of the AI chip market
• According to research: NVIDIA holds 80% of the AI chip market, with a market cap of $4.55 trillion, Q3 2025 revenue of $57 billion, and $500 billion in unfulfilled orders
• According to research: Google’s TPU is difficult to disrupt NVIDIA due to its closed ecosystem (tied to Google Cloud) and lack of CUDA support

• Reddit user: Google’s TPU+OCS architecture has significant advantages but still relies on NVIDIA GPU flexibility; Anthropic’s million-TPU order validates demand, optimistic about optical modules (LITE), domestic Xuchuang/Shenghong
• Snowball user: Google’s technology is strong but TPU lacks CUDA ecosystem; AI is not a bubble (dynamic computing vs static software), negative for low-bargaining-power software companies, positive for hardware and energy infrastructure

Both recognize the core status of NVIDIA’s ecosystem barriers; Google’s TPU is a niche technology driving supply chain growth; investment focus is on computing power hardware, energy storage, and storage sectors