Do Not Rely on Chinese AI, Says Lux Capital's Josh Wolfe

Key Concepts:

• AI competition between the US and China
• Chip restrictions and their impact on innovation
• Open source vs. closed source AI models
• Data as a competitive advantage in AI
• The role of use cases in AI dominance
• Grand bargain with China (trade, tariffs, and other issues)
• Investment restrictions in Chinese technology companies
• Impact of research funding cuts on innovation
• Bayh-Dole Act and its role in commercializing research

AI and China: A Wake-Up Call

• Chip Restrictions and Innovation: The initial US strategy focused on restricting chip exports to China. However, Josh Wolfe points out that "necessity is the mother of invention," suggesting that these restrictions may inadvertently spur Chinese innovation in chip design and alternative architectures.
• DeepSeek as an Example: DeepSeek's ability to achieve significant results with fewer GPUs served as a "wake-up call." It raises questions about the future of AI, potentially favoring smaller models and different chip architectures.
• Open Source vs. Closed Source: Wolfe believes that while large companies like OpenAI and Anthropic have raised billions, the future advantage lies with companies possessing "longitudinal silos of data." These companies can leverage free, open-source models on their proprietary data. Examples include:
  • Pharma companies (clinical trial data)
  • Bloomberg (financial information)
  • Meta (WhatsApp, Instagram, Facebook data)
  • Tesla/X (Twitter posts)
• Focus on Applications and Data: Wolfe argues that the US should shift its focus from competing on chips and models to competing on the "applications and the use of the data to actually do stuff."
• China's AI Capabilities: Jensen Huang believes China is very close to the US in AI capabilities.
• Importance of Open Models: Wolfe emphasizes the need for the world to use open AI models developed by companies that prioritize truth and avoid censorship, contrasting this with Chinese models that may be subject to political constraints (e.g., censorship of Tiananmen Square, Xinjiang, and Uyghur issues).

Trade and Tariffs: The Grand Bargain

• Tariff Policy: Wolfe suggests that the current tariff policy could be part of a "grand bargain" with China, encompassing various issues beyond trade imbalances.
• Potential Components of the Grand Bargain:
  • TikTok
  • Fentanyl
  • IP theft and security
  • Human rights issues
  • Environmental issues
  • Freedom of navigation in the South China Sea
• Volatility and Unpredictability: The initial impact of the policy has been characterized by volatility and surprise, reminiscent of the Trump 1.0 era.
• Impact on Small Businesses: High tariffs could lead to bankruptcies for small businesses, creating pressure for alternative solutions.
• Friend-shoring: Wolfe suggests "friend-shoring" across various countries as a potential solution.

Investment Restrictions and National Security

• Red Line: Wolfe believes that venture capital firms should not invest in Chinese technology companies that are part of China's "military-civil fusion."
• Control on Capital: He advocates for a ban on investments in technologies that could be used as weapons systems or against the US.
• Insidious Threats: Wolfe emphasizes that there are "a lot of threats from China that are more insidious than people understand."

Research Funding and Innovation

• Impact of Research Cuts: Cuts to research funding could have lasting damage on new businesses and drug treatments.
• The Bayh-Dole Act: Wolfe explains the ecosystem where taxpayer dollars flow from the government to agencies like the NSF and NIH, then to universities and researchers. The Bayh-Dole Act allows universities to spin out patents and license intellectual property to companies. This system has been crucial for creating companies like Google, Microsoft, and Genentech.
• Commercialization Opportunities: Despite potential funding cuts, Wolfe's firm is more excited than ever to capitalize on scientific work from universities and government labs, even at earlier stages of development (e.g., 60-70% "baked" instead of 80%).

Conclusion:

Josh Wolfe presents a multi-faceted view of the US-China competition, particularly in AI. He argues that focusing solely on chip restrictions is insufficient and that the US needs to leverage its data advantage, promote open AI models, and address broader issues through a "grand bargain" with China. He also stresses the importance of protecting national security by restricting investments in Chinese technologies with military applications and maintaining a robust research funding ecosystem to foster innovation. The key takeaway is a shift in strategy from direct competition on hardware to leveraging data, applications, and strategic alliances.