Renowned Analyst: AI-Driven Semiconductors Entering First True Demand Supercycle
Taylor Wilson
Bernstein senior semiconductor analyst Stacy Rasgon says AI is pushing chips into the first genuine demand supercycle in his twenty-year career — from GPUs to networking gear, the entire chain is stretched, and the only consensus is: nobody has enough compute.
Why does "supercycle" actually mean something this time?
Over the past two decades, semiconductors have been labeled a "supercycle" multiple times. Rasgon argues every prior instance was a pulse in a single product category or a localized demand spike.
This means → past "supercycles" looked more like one bottleneck tightening up, not the whole chain running short at once.
This time, AI-driven demand is pulling on chips, memory, optical modules, and networking infrastructure simultaneously — the reason Rasgon says the word finally fits.
Which links in the chain are being pulled at once?
AI compute demand is lifting GPUs, high-bandwidth memory (HBM), optical transceivers, networking equipment, and CPUs — not one component in shortage, but the full stack under pressure together.
In plain terms = before, one dish ran out; this time the entire table is short.
Rasgon notes the industry's only consensus right now: nobody has enough compute. That single sentence tells you demand has overshot every player's forecast.
How long can this cycle last — and what decides it?
Whether the supercycle holds depends on one core variable: can the supply gap narrow in the near-to-medium term?
This means → if capacity expansion cannot keep pace with AI compute demand growth, the gap widens and the cycle's staying power strengthens.
Conversely, once capacity floods in and demand growth slows, the cycle could peak early. The timing mismatch between supply and demand is the key signal for the turning point.
Content is for reference only, not financial advice.