Dylan Patel – Anthropic & OpenAI will have most of the world’s compute by 2028

Anthropic and OpenAI are projected to control the majority of the world’s compute by 2028 due to their superior ability to monetize compute resources and outbid competitors. This centralization is driven by economies of scale in training, compute scarcity, and advancements like continual learning and Recursive Self-Improvement (RSI). The immense capital expenditure required for AI infrastructure raises concerns about potential sovereign debt crises and the impact on non-AI exposed economies and equities.

• Anthropic and OpenAI are expected to control most of the world’s usable compute within a few years.
• The shift from inference to training is accelerating as RSI approaches.
• Massive AI capital expenditure, potentially exceeding $10 trillion by 2030, could trigger a sovereign debt crisis.
• Hyperscaler debt may drive up interest rates, leading to bankruptcies in non-AI countries and a crash in non-AI equities.
• Forces driving centralization in the AI industry include economies of scale in training, compute scarcity, and continual learning.
• The ability of labs to monetize compute better than others allows them to outbid competitors for resources.
• China currently accounts for less than 10% of new compute deployment.
• There is debate on whether anything can counter the trend towards AI industry centralization.
  Continue reading https://www.dwarkesh.com/p/dylan-patel-3