Dylan Patel - Inside the Trillion-Dollar AI Buildout - [Invest Like the Best, EP.442]

NVIDIA funds OpenAI's compute purchases (of NVIDIA chips) with an equity investment. This effectively gives OpenAI a discount without lowering market prices, while NVIDIA gains equity in a key customer and locks in massive sales.

The massive capital investment in AI infrastructure is predicated on the belief that more compute will always lead to better models (scaling laws). If this relationship breaks, the glut of data center capacity will have no ROI, triggering a severe recession in the tech and semiconductor sectors.

The guest argues that without the massive GDP growth and efficiency gains promised by AI, the U.S. is on a path to being surpassed by China as the world hegemon by 2030. AI is not just an economic boom; it's a geopolitical necessity for maintaining America's global standing.

A 10x increase in compute may only yield a one-tier improvement in model performance. This appears inefficient but can be the difference between a useless "6-year-old" intelligence and a highly valuable "16-year-old" intelligence, unlocking entirely new economic applications.

Paying a single AI researcher millions is rational when they're running experiments on compute clusters worth tens of billions. A researcher with the right intuition can prevent wasting billions on failed training runs, making their high salary a rounding error compared to the capital they leverage.

The insatiable demand for power from new data centers is so great that it's revitalizing America's dormant energy infrastructure. This has led to supply chain booms for turbines, creative solutions like using diesel truck engines for power, and even a doubling of wages for mobile electricians.

While OpenAI pursues a broad strategy across consumer, science, and enterprise, Anthropic is hyper-focused on the $2 trillion software development market. This narrow focus on high-value enterprise use cases is allowing it to accelerate revenue significantly faster than its more diversified rival.

Pre-training on internet text data is hitting a wall. The next major advancements will come from reinforcement learning (RL), where models learn by interacting with simulated environments (like games or fake e-commerce sites). This post-training phase is in its infancy but will soon consume the majority of compute.

The next human-computer interface will be AI-driven, likely through smart glasses. Meta is the only company with the full vertical stack to dominate this shift: cutting-edge hardware (glasses), advanced models, massive capital, and world-class recommendation engines to deliver content, potentially leapfrogging Apple and Google.

The US and China have divergent AI strategies. The US is pouring capital into massive compute clusters to build dominant global platforms like ChatGPT (aggregation theory). China is focusing its capital on building a self-sufficient, domestic semiconductor and AI supply chain to ensure technological independence.

Companies like OpenAI and Anthropic are intentionally shrinking their flagship models (e.g., GPT-4.0 is smaller than GPT-4). The biggest constraint isn't creating more powerful models, but serving them at a speed users will tolerate. Slow models kill adoption, regardless of their intelligence.

The traditional SaaS model—high R&D/sales costs, low COGS—is being inverted. AI makes building software cheap but running it expensive due to high inference costs (COGS). This threatens profitability, as companies now face high customer acquisition costs AND high costs of goods sold.