Inference Spend Isn't Speculative: Why AI Tokens Differ From Dot-Com Hardware Hoarding
AccBalanced · x · 2026-08-26
Neil Movva, a former Nvidia engineer, argues that current AI spending differs fundamentally from the dot-com bubble:
- Historical Context: Past investments in networking equipment were speculative bets on future users that never materialized.
- Token Consumption: People buy tokens for immediate utility, not hoarding, eliminating speculation.
- Training vs. Inference: While the recent chip shortage was driven by speculative training spend, inference spend is inherently non-speculative and monotonically increasing.
Neil runs Sail Research, building infrastructure to make tokens cheaper for agents.
More from Venture
- Spending $12.7k for a #1 Spot Generated $600k Annual Pipeline — jonathan_wilke · 2026-08-26
- Investor Lists Top Picks: Robotics, AI Models, and Chips — vaibhavbetter · 2026-08-26
- Skepticism on Anthropic's $30T TAM: calculating ROI in a winner-takes-all market — teortaxesTex · 2026-08-26
- a16z Deck: Intelligence is the Primitive, Applications are the Diffusion Layer — illscience · 2026-08-26
- Polymarket: 48% Chance OpenAI IPO Exceeds $1.5T Valuation — Polymarket · 2026-08-26
- Icon keeps the human in UGC ads: real creators, $999 for 6 finished spots — aryanXmahajan · 2026-08-26