Why The AI Bubble May Be Good

Quick Overview

The AI bubble, characterized by massive investments and circular deals among major tech companies, is argued to be rational because it reflects a fundamental societal need for ever-increasing information processing capacity, even if it creates system-wide vulnerability to an AI crash, as demonstrated by Google's strategic shift toward in-house hardware like TPUs to reduce reliance on NVIDIA.

Key Points: The current AI boom involves circular investments where AI companies borrow money to invest in other AI companies, which the speaker analogizes to diversifying a stock portfolio to manage risk. Google is actively competing with NVIDIA by pushing its specialized Tensor Processing Units (TPUs) like the TPU v4, which offer better performance per watt and lower cost due to in-house co-design. Google's TPU v7 offers roughly 2.8x better performance per watt than NVIDIA's H100 and beats newer Blackwell GPUs in energy efficiency, translating to millions saved on electricity and cooling. Anthropic signed a $52 billion deal to purchase Google's TPU v7 chips, marking a defection from NVIDIA's ecosystem, which typically charges high margins (70-80%). The massive capital flowing into data centers (estimated in the trillions of dollars) is driven by the demand for computation, not just building infrastructure, highlighting the focus on processing power. The speaker argues that while the AI industry dependency creates systemic risk (vulnerability to an AI crash), the circular investments simultaneously provide diversification against the failure of any specific AI company.

Context: The video features Dr. Sabine Hossenfelder discussing the financial dynamics surrounding the current Artificial Intelligence (AI) boom, specifically addressing whether the heavy, interconnected investments—often referred to as 'circular deals'—constitute an unsustainable bubble. She contrasts the high capital expenditure, particularly in hardware like GPUs and TPUs, with historical investment bubbles like the 17th-century Tulip Mania, while also examining recent strategic moves by major players like Google and Anthropic to diversify their reliance on dominant suppliers like NVIDIA.