What If AI Is a Bubble?

(theatlantic.com)

2 points | by paulpauper 12 hours ago ago

3 comments

  • iopjgalejandro 11 hours ago

    The article's "reverse terraforming the Earth" quote is the real story here. This is what makes the dot-com analogy imperfect.

    The dot-com bubble required relatively little CapEx. This AI bubble is a physical infrastructure bubble (data centers, power, $5T in chips).

    It feels far more analogous to the fiber-optic bubble of the late 90s. Trillions were spent laying "dark fiber" across the globe. The companies that laid it all went bust, but that exact infrastructure was what enabled the next generation (Google, Facebook, etc.) to be built for almost free a few years later.

    We are in the "building the dark fiber for AI" stage. The immediate ROI isn't there, but the infrastructure is the prerequisite.

    • thelastgallon 8 hours ago

      GPUs depreciate very very fast, unlike compute/network. Nvidia is coming up with newer, significantly better products very fast. These should be depreciated over 2 - 3 years, but they are being depreciated over 6 years. Michael Burry is shorting based on this: https://www.youtube.com/shorts/7hIcgCz1vpQ

    • techblueberry 11 hours ago

      By infrastructure do you mean GPU’d or models?

      “ Dark fiber, as an infrastructure, is designed to last for decades. It is common for dark fiber to be installed with a lifespan of no less than 20 years, and in many cases, it can remain functional for up to 30 years or more without degradation.”

      I dunno, what if it’s all for naught? What if it’s more like crypto and becomes a medium sized niche technology?

      “ global cryptocurrency market size was estimated at USD 5.70 billion in 2024 and is projected to reach USD 11.71 billion by 2030, growing at a CAGR of 13.1% from 2025 to 2030. The increasing adoption of distributed ledger technology is anticipated to propel the cryptocurrency industry growth during the forecast period.”