The Circuit EP 143: NVIDIA Earnings! Google and Gemini TPUs FTW
39 snips
Nov 24, 2025 The conversation dives into NVIDIA's impressive earnings and what it means for the market. They explore the dynamics of cloud compute pre-purchases and NVIDIA's training strategies. The discussion shifts to the rapid evolution of AI-optimized data centers and the power of Google's TPUs. Ben shares exciting insights on NanoBanana's Notebook LM, which revolutionizes infographic creation from data and handwritten notes. They also touch on productivity implications and how multimodal experiences can drive broader user adoption.
AI Snips
Chapters
Transcript
Episode notes
GPU Depreciation Is An Accounting Tension
- Depreciation debates stem from the gap between accounting life and economic utility of GPU servers.
- Rapid annual product cadence forces obsolescence pressure and complicates asset valuation.
GPU Spot Prices Reveal Three-Year Economics
- Spot pricing shows elasticity across GPU generations, with newer Blackwell instances commanding far higher rates.
- Price curves imply an approximate economic GPU life of three years for profitability on many workloads.
Build Purpose-Built AI Data Centers
- Hyperscalers should purpose-build AI-optimized data centers to gain performance and efficiency advantages.
- Invest in greenfield AI factories and tuned stacks rather than retrofitting general-purpose racks.