Is the AI market driven by fundamental value or just the "Pink Elephant" of FOMO and herd mentality?
In this episode, Paul Hammer and Tony Grayson strip away the hype to discuss the brutal economics of the AI infrastructure build-out. From Satya Nadella’s strategy of using Neoclouds as a hedge to Werner Vogels’ assertion that the cloud is now purely industrial, we explore why the industry must shift its focus from raw power to Tokens per Watt.
We dive deep into the looming risks of 2027, the impact of 18-month hardware refresh cycles on balance sheets, and why the current "gigawatt or bust" mentality might lead to a massive market correction.
Key Topics Covered:
The "Pink Elephant": How market psychology and recency bias are driving a repeat of the dot-com bubble.
The Winner's Curse: Why Microsoft is de-risking and why CoreWeave might be left "holding the bag" on H100s.
Unit Economics: Why "Tokens per Second" and "Tokens per Watt" are the only metrics that matter for profitability.
Hardware Cliffs: The financial danger of treating data centers as 18-month assets and the risk of massive write-offs (up to $53bn).
Design & Power: Why H100 data center designs don't map to Grace Blackwell or Vera Rubin, and the reality of SMR nuclear timelines vs. marketing fluff.
AEO vs. SEO: How Answer Engine Optimization is changing how we create content for Gemini, Claude, and ChatGPT.
Podden och tillhörande omslagsbild på den här sidan tillhör
Uptime Punks. Innehållet i podden är skapat av Uptime Punks och inte av,
eller tillsammans med, Poddtoppen.