Future of AI Demand, as Nvidia CEO Slams US Chip Policy

Key Concepts:

• AI Accelerators
• GPUs (Graphics Processing Units)
• Large Language Models (LLMs)
• Training vs. Inferencing
• Sovereign AI
• CapEx (Capital Expenditure)
• Hyperscalers

China's AI Market and Nvidia's Position

• China is a world leader in AI accelerators.
• Companies like AMD cannot fully access the Chinese market, where significant AI development and data center construction are occurring.
• Nvidia GPUs are still used for training LLMs within China, such as Deep Sea and Alibaba's models.
• Nvidia's inability to sell to China, which accounts for approximately 20% of the $250 billion global GPU spend, means they lack a "sovereign AI" opportunity in that region.
• Nvidia has secured deals in the Middle East.

Global LLM Landscape

• The US and China lead in LLM development.
• The top four or five LLMs are based in the US, with the remaining prominent models in China.
• Other countries are investing in infrastructure but lack the same level of development as the US and China.

Deep Sea and the Shift from Training to Inferencing

• Deep Sea's emergence at the beginning of the year forced companies to re-evaluate their upfront spending on pre-training.
• This led to a pivot towards reasoning and inferencing, as seen with OpenAI's models (e.g., one and three) and Google's 2.5 Pro.
• The thesis presented on March 17th highlighted a shift from training to inferencing.
• The inferencing opportunity is expected to grow faster and become larger than the training market.
• The overall AI market will continue to grow, but the balance will shift towards inferencing.

CapEx for Hyperscalers

• The research mentioned a $525 billion CapEx for hyperscalers by 2032 in the context of the Nvidia story.

Conclusion

The AI market is evolving, with a significant shift from training to inferencing. While Nvidia faces challenges in accessing the Chinese market, the overall AI market is expected to grow, driven by the increasing importance of inferencing. The US and China are the leading players in LLM development, with other countries investing in infrastructure but lagging behind in model prominence.