AI Fund’s GP, Andrew Ng: LLMs as the Next Geopolitical Weapon & Do Margins Still Matter in AI?

Core Bottlenecks: Andrew Ng identifies electricity, semiconductors, data, and algorithms as the four key parameters for AI, with electricity and semiconductors currently being the most pressing short-term constraints.

• Electricity Constraints: Many US and Western data center operators are stalled by permitting delays and local community opposition, despite data centers being critical infrastructure for the digital economy.
• China's Advantage: In contrast to the West, China is aggressively building power plants, including nuclear facilities, to support its industrial AI commitment.
• Compute Insatiability: Ng states he has never met an AI professional who feels they have enough compute; demand for inference tokens consistently outstrips supply, causing frustration for users of cloud coding and generation tools.

Geopolitical Dynamics & Open Models:

• China's Open Strategy: China is leading in releasing open-weight models, which Ng views as a strategic advantage for knowledge circulation and innovation within their economy.
• Soft Power: Open-weight models are a significant source of geopolitical influence, as they allow nations to export their values and narratives to developing markets globally.
• Export Control Backfire: Ng argues that US export controls on semiconductors (e.g., to Huawei and NVIDIA) have inadvertently accelerated China's domestic semiconductor development, incentivizing them to build competitive offerings using less powerful chips in larger clusters.

Workforce Transformation & Hiring Trends:

• The "10x" Engineer: The most productive engineers are currently those with 10-20 years of experience who have effectively integrated AI into their workflow, surpassing fresh graduates who are still coding in the pre-AI style.
• The Vulnerable Cohort: The group most at risk is fresh college graduates who have learned coding but lack AI proficiency, a gap exacerbated by slow university curriculum updates.
• Hiring Shifts: Companies are moving away from hiring fresh grads who only know traditional coding; Ng notes he no longer hires experienced developers who refuse to adapt to AI tools.
• Coding Democratization: Ng advocates for teaching all students to code, arguing that the ability to direct AI via code empowers non-engineers (e.g., marketers building their own apps) to execute projects that were previously impossible.

Economic Impact & Investment Outlook:

• GDP Growth Potential: Ng expects AI to drive GDP growth of 5-6% or higher by democratizing expensive intelligence (like medical advice and tutoring), contrasting with Andrej Karpathy's more conservative 2% growth estimate.
• Value Realization: The primary path to value is not simple cost savings but re-engineering workflows to either "do more" (scale to new markets) or "do faster" (reduce turnaround times, e.g., loan approvals from weeks to minutes).
• Application Layer Investment: While infrastructure investment is clear, the application layer presents a challenge for VCs to deploy large capital sums due to the low cost of building initial prototypes.
• Margin Realities: While AI startups often operate with low margins today due to high token costs, Ng forecasts that falling token prices will allow efficient operators to bend cost curves back down rapidly.

Agentic Workflows & Defensibility:

• Utility of Agents: Ng disagrees with the notion that useful AI agents are a decade away, citing existing workflows in tariff compliance, medical assistance, and legal document processing that would be impossible without agentic systems.
• Moat Evolution: The "software moat" (based on code complexity) is weakening as AI lowers the barrier to entry, but new moats are forming around brand reputation, consumer trust, and two-sided network effects.
• Enterprise Adoption: The biggest barrier for enterprise AI is not data but change management and organizational culture, with adoption likely taking longer than current hype suggests.

Policy & Regulation:

• US Regulatory Landscape: Ng supports the removal of unnecessary regulations and the Schumer AI Insight Forum's focus on investment over stifling anti-competitive rules.
• Talent Attraction: He warns that failing to attract global talent (including high school and university students) due to restrictive policies would be a significant "unforced error" for the US.
• Europe's Challenge: Ng urges European regulators to stop focusing on leading in AI regulation and instead focus on investing and building, warning that over-regulation acts as a competitive disadvantage.

Operational Philosophy & Personal Views:

• AI Fund Model: Ng's organization operates as a venture studio rather than a traditional fund, co-founding companies with founders from the idea stage rather than just investing in existing startups.
• Work Ethic: Ng defends a strong work ethic and the "996" intensity seen in China, arguing that hard work remains essential for building a dent in the universe, while acknowledging the need for respect for those unable to work such hours.
• Hype Concerns: Ng expresses concern that doomsday hype regarding human extinction is deterring young talent from pursuing AI careers, which hinders the workforce's ability to transition.

Forward-Looking Statements:

• Tech Evolution: Ng anticipates a future with a diverse ecosystem of models ranging from tiny, locally run models for simple tasks to massive models for complex reasoning, rather than a single monolithic architecture.
• Decade-Long Growth: He predicts that AI capabilities will continue to improve meaningfully for at least a decade, with the industry still identifying valuable applications and building infrastructure ten years from now.
• Medical Breakthroughs: Ng cites the potential for incredible medical discoveries, particularly in treating diseases like his mother's Multiple Sclerosis, as a primary source of future optimism.