July 21, 2026
Chips, lies, and power bills
The Growing Compute Shortage
AI’s new gold rush has everyone fighting over chips, power, and who gets blamed
TLDR: AI companies are racing for the hardware and electricity needed to run smarter tools, and supply is struggling to keep up. In the comments, readers fought over whether this is a true shortage or just sky-high prices, while others roasted the charts and joked that maybe AI is already training us.
The big panic in AI right now is simple: there may not be enough machine power to go around. The article argues that demand for AI is exploding so fast that the physical stuff behind it—chips, memory, factories, and even electricity—is getting squeezed all at once. Translation for normal humans: AI may look like magic on a screen, but behind the scenes it’s chewing through expensive hardware and huge amounts of power, especially as newer AI helpers do more steps, make more attempts, and burn far more resources than old-school chatbots.
But the comments? That’s where the sparks flew. One camp wanted better proof, with readers dragging the article’s charts for bad scaling and missing context. One especially savage reaction basically called the graph a statistical crime scene, while another politely asked for a simple global chart showing how much data-center capacity exists and what’s coming next. Then came the philosophy-brain crowd: one commenter joked that maybe the real story is not humans training AI, but AI training humans to serve its goals. Casual nightmare fuel!
And of course, there was classic internet contrarian energy. While the article screams shortage, one commenter shot back: GPUs aren’t really scarce, they’re just absurdly expensive—which is a very Silicon Valley distinction. Another reader went full sci-fi, daydreaming about AI compute satellites and the first baby steps toward a giant space brain. So yes, the infrastructure crunch is real, but the community is split between chart police, doom prophets, rich-guy pragmatists, and starry-eyed futurists.
Key Points
- •The article argues that AI compute is becoming a scarce strategic resource because demand is rising faster than supply across the infrastructure stack.
- •Agentic AI systems, especially coding agents, are described as far more compute-intensive than traditional chatbots, consuming 100x to 1,000x more tokens in some cases.
- •The article says shortages are emerging simultaneously in deployed GPUs, advanced semiconductor manufacturing capacity, memory supply, and power infrastructure.
- •TSMC’s advanced-node capacity, particularly N3, is described as nearing full utilization through at least 2027, limiting expansion of AI accelerator production.
- •High-bandwidth memory remains undersupplied, while broader DRAM prices have risen sharply as manufacturers shift capacity toward HBM, affecting other electronics industries as well.