# US vs China: Who Is Actually Winning The Chip War? (Jensen Huang Scandal)

## Executive summary

The global AI landscape is defined by a compute constraint, driven by geopolitical tensions surrounding advanced chips like NVIDIA's GB300. While an ongoing investigation into Jensen Huang's personal life and business dealings adds uncertainty, the core issue remains the rapid acceleration of AI research (recursive self-improvement) and the critical need for advanced compute power for both commercial deployment and cyber defense. Companies must plan for limited access to frontier models, making compute availability an existential factor for future AI development.

## Key takeaways

- Geopolitical Chip Bottleneck: Beijing is reportedly holding up approximately 600,000 H200 chips in Hong Kong, with the reluctance attributed to a desire to reduce dependence on American suppliers like NVIDIA. This creates a complex market where Chinese companies may prefer renting advanced equipment overseas rather than relying on direct shipments.
- AI Research Acceleration: AI research is undergoing 'recursive self-improvement,' where models use AI to write research code and prepare experiments, leading to rapid progress. METR assessed that Anthropic's models are improving roughly twice as fast as between Spring 2024 and Fall 2025.
- The Investigation's Impact: An ongoing investigation into NVIDIA's board is examining whether Jensen Huang's personal relationship inappropriately influenced who received access to scarce advanced chips. The outcome of this investigation could significantly affect the speed of AI development in both American and Chinese companies.
- Cyber Defense and Compute: The theft of $6.34 billion from Binance demonstrated how attackers leveraged internal AI tools. This highlights that access to frontier intelligence and robust compute is now critical for cyber defense, making compute availability a national priority.

## Technical details

- Chip Generation Comparison: The difference between chip generations is critical for AI capacity. An H200 is an older NVIDIA processor used for training and running large AI models, while GB300 systems utilizing the newer Blackwell Ultra generation offer significantly greater capacity for demanding AI work.
- Advanced Compute Access (Mega Speed): Mega Speed, a Singapore-based company, rents advanced NVIDIA computing power (including GB300 processors) from data centers in Southeast Asia to international clients like Alibaba. This demonstrates how advanced compute can be accessed indirectly, even when the client is in a different country.
- AI Research Methodology: AI labs are increasingly using coding agents and AI tools to write research code and prepare experiments. This 'assisted recursive self-improvement' allows researchers to accelerate progress on specific abilities, even if fully autonomous self-improvement is still distant.

## Practical implications

- Companies must plan for compute constraints, assuming that access to advanced chips will be difficult and highly regulated.
- Cyber defense strategies must account for the increasing sophistication of AI-powered attacks, requiring robust, secure internal AI tools.
- Long-term AI planning (e.g., 2027 budget cycles) must factor in the potential for geopolitical restrictions on chip availability and deployment.
- Security teams must focus on the scope and security of AI agents, as internal tools can be compromised and used for malicious purposes.

## Topics

Artificial Intelligence (AI), Semiconductors/Hardware, Geopolitics, Cybersecurity, Compute Infrastructure, Nate's Library MCP, Alibaba, Mega Speed, Anthropic, Binance

Source: https://www.youtube.com/watch?v=OEY-vm7qRHc
