Pourquoi la Silicon Valley abandonne Nvidia ?

The video presents a compelling narrative about the competitive dynamics in AI hardware, focusing on the shift from Nvidia GPUs to Google’s TPUs. The analysis is well-structured, starting with Nvidia’s record earnings and contrasting them with a stock drop, which sets up the central question. The presenter uses specific examples, such as Anthropic’s $200 billion potential deal and Midjourney’s 65% cost reduction, to illustrate the economic incentives for migration. These examples are effective in making the argument tangible. The technical explanation of the systolic array and the historical context of the TPU’s development in 2013 add depth. However, the video lacks direct citations to primary sources; the claims about market share and deals are presented as facts without referencing specific reports or announcements. The presenter’s own involvement with Antimatter (a company in the AI infrastructure space) introduces a potential conflict of interest, though it is disclosed. The argumentation is logically sound, but the reliance on anecdotal evidence and the absence of counterarguments (e.g., Nvidia’s software ecosystem moat) weakens the critical analysis. The video does not address potential drawbacks of TPUs, such as vendor lock-in with Google Cloud. Overall, the video is informative and engaging, but its credibility is limited by the lack of transparent sourcing and the presenter’s commercial interests.

The title accurately reflects the content, which focuses on why major AI companies are diversifying away from Nvidia GPUs to custom chips like Google's TPU.

The video provides a well-structured analysis of the competitive dynamics in AI hardware, citing specific deals and technical concepts. However, it lacks direct citations to primary sources and relies on the presenter's expertise and industry reports. The analysis is coherent but could benefit from more transparent sourcing.

• 00:00Introduction: Nvidia's record earnings and stock drop.
• 02:00Google announces TPU v8t and v8i at Las Vegas event.
• 04:00Anthropic, Meta, and OpenAI start using Google TPUs.
• 06:00Nvidia's market share projected to fall from 90% to 70%.
• 08:00Midjourney case study: 65% cost reduction after migrating to TPUs.
• 10:00History of TPU development: Jeff Dean's 2013 calculation.
• 12:00Explanation of systolic array architecture.
• 14:00Comparison of GPU vs TPU efficiency.
• 16:00Discussion of Nvidia's CUDA software moat.
• 18:00Conclusion: Who will win the AI chip war?

• Newsletter Grand Angle Nova ✓ verified — Promotional link for the channel's newsletter.

• Nvidia Q4 2025 earnings report — Reported record revenue of $81.6B, but stock fell due to future competition concerns.
• Google TPU v8 announcement — Google unveiled TPU v8t and v8i at an event in Las Vegas.

• Nvidia's CUDA ecosystem — Some argue that Nvidia's software moat (CUDA) remains a significant barrier to switching, which the video downplays.