Particle.news
Download on the App Store

AI Providers Ration Access as a Compute Crunch Takes Hold

Compute limits are forcing a turn from flat plans to metered use.

Overview

  • New reporting describes an industrywide compute crunch as providers tighten usage, with GitHub pausing Copilot signups and removing higher‑cost models.
  • Anthropic curtailed high‑load agent traffic, cut off OpenClaw users, tested removing Claude Code from its $20 plan, and pushed users toward pay‑as‑you‑go API billing.
  • Tokens, the unit that measures and bills model usage, are piling up under always‑on AI agents, which makes flat subscriptions hard to sustain.
  • Executives and analysts point to a shortage of computing power, with reports noting OpenAI’s CFO has said the company does not have enough compute.
  • Gartner estimates the business will need close to $2 trillion a year by 2029 or 50,000–100,000x more token use by 2030, while parts prices, power bills, and local resistance to new data centers show the strain spilling into daily life.