Wired's Steven Levy on DeepSeek's latest AI model, state of AI tech race
Speakers
Steven Levy
— Editor at Large, Wired
Summary
Steven Levy joined Squawk Box to discuss DeepSeek's expected new model, its more efficient training approach, and what it could mean for AI spending and Nvidia chip demand. He argued large language models are commoditizing and that value is shifting to applications and developer networks. Levy also assessed Microsoft, Google, Meta, Apple, and xAI/X, highlighting AI productization advantages and challenges.
- DeepSeek's next model is expected around Lunar New Year and may have strong coding abilities.
- DeepSeek's paper describes more efficient LLM training needing fewer chips, raising AI-spending questions.
- Nvidia chip export restrictions and possible smuggling to China remain a focus.
- Levy sees LLM commoditization pushing value toward applications and developer networks.
- Microsoft and Google are positioned well through productivity and search distribution.
- Meta is behind in AI productization but trying to build its way out.
- Apple faces AI integration and CEO succession uncertainty.
- xAI is a wild card; X content moderation concerns were flagged.
Ideas
DeepSeek efficiency could pressure AI chip demand.
DeepSeek's reported new training method is more efficient and requires fewer chips, which could become another industry-wide shock that raises doubts about excessive AI compute spending. However, DeepSeek may still be using restricted top-end Nvidia chips, so this is a setup to monitor for Nvidia and AI compute demand rather than a clear bearish call.
AI app layer is key differentiator.
Large language models are commoditizing, so the winners will be companies that translate models into useful work by building applications and developer networks rather than only providing raw models.
Microsoft has strong AI distribution channels.
Microsoft has the productivity software installed base and OpenAI partnership, and is hedging by looking at other large language models, giving it strong distribution to embed AI into useful work.
Google search dominance aids AI productization.
Google's search dominance gives it a powerful built-in platform to distribute and productize AI applications as models commoditize.
Meta is behind in AI productization.
Meta is behind in AI productization because AI is not core to its connect-the-world roots and users are less eager to use AI to improve Instagram, though the company is trying to build its way out.
Apple faces AI and succession uncertainty.
Apple has struggled to integrate AI into its products, and there is no clear successor pegged for Tim Cook amid retirement speculation, creating execution and leadership uncertainty.
