The World Next Week New Podcast Spotlight: The Interconnect
Feb 14, 2025
Mark Horowitz, Chair of Electrical Engineering at Stanford, and Sebastian Elbaum, a tech expert from the Council on Foreign Relations, dive into the future of chip manufacturing and its impact on AI. They highlight challenges to Moore's Law and the geopolitical implications of semiconductor tech, particularly U.S. export restrictions on China. The guests also discuss energy challenges for AI and brainstorm whimsical app ideas, blending serious topics with creative humor. This engaging dialogue connects the dots between technology and policy like never before.
AI Snips
Chapters
Transcript
Episode notes
Moore's Law and Cost Scaling
- Moore's Law, stating that transistor count doubles roughly every two years, has held true for a long time.
- However, the associated cost scaling has dramatically changed; cost per transistor isn't decreasing exponentially anymore.
Electronics and Computation Costs
- Falling computation costs incentivized moving functionalities into the computation domain.
- This was driven by both cost-effectiveness and the ease of adding new features through electronics.
DeepSeek's AI Model
- Chinese startup DeepSeek developed AI models with performance comparable to US companies but at much lower training costs.
- This was achieved through clever optimizations due to constraints in accessing high-end hardware like NVIDIA chips.