OpenAI’s ChatGPT, Codex and API Suffer Hours-Long Global Blackout on 20 April 2026

Centralized tech backbones have buckled before—AT&T’s long-distance network crash on 15 Jan 1990 silenced 60,000 calls, and Amazon’s 28 Feb 2017 S3 failure darkened a third of the commercial web—but this is one of the first times an AI utility with ~200 million daily users went dark. The event highlights a structural tension: large-language models deliver economies of scale yet create single points of failure, echoing 20th-century electric-grid debates that led to redundancy standards. As software workflows, customer support, even homework become algorithm-mediated, reliability expectations will migrate from ‘best-effort website’ to ‘critical infrastructure.’ Whether OpenAI’s still-undisclosed root cause leads to architectural changes or is treated as routine will shape public trust and regulatory appetite over the next decade—and, on the century horizon, determines whether AI platforms evolve like robust public utilities or brittle proprietary silos.