According to an analysis by Ookla of 471 days of Downdetector data in the United States, representing about 3.7 million reports across major AI platforms (ChatGPT, Claude, Gemini, Copilot) plus AWS and Azure, major outage days on the AI app side have multiplied by 8 in one year. There were 6 such days in Q1 2025, and 51 in the same quarter of 2026. The worst offender was Claude, accounting for 39 of those 51 bad days, compared to 7 for Gemini, 3 for Copilot, and 2 for ChatGPT. In March 2026 alone, Claude racked up almost 3 times more reports than in February. Conversely, ChatGPT saw its monthly median error rate drop between April 2025 and April 2026. The analysis notes that outages often stem from the product layer, cloud hosting, or access layer, exemplified by a major AWS outage on October 20, 2025, where the DNS management system went haywire, and an Azure crash 9 days later.