On April 23, 2026, xAI’s artificial intelligence assistant, Grok, experienced a series of intermittent service outages that left thousands of users unable to access the platform for several hours. Reports of connectivity issues began surfacing at approximately 08:30 UTC, with the frequency of errors peaking during mid-day trading hours in North America. Users attempting to interact with the Grok-3 model were met with persistent error messages stating that the service was at capacity or experiencing high demand.
The disruptions primarily affected subscribers of the X Premium and Premium+ tiers, who have exclusive access to the chatbot. Internal data logs and third-party monitoring services indicated that API response times for Grok-3 spiked from a standard 200 milliseconds to over 15 seconds before the service eventually timed out. Technical analysts attribute the instability to the massive influx of concurrent users following the recent rollout of multimodal capabilities, which allow Grok to process real-time video and audio data. This increased computational load appears to have strained the load-balancing protocols within xAI’s primary data centers.
xAI has been aggressively scaling its hardware footprint, recently completing the expansion of its Colossus supercomputer cluster to include over 200,000 Nvidia Blackwell B200 GPUs. Despite this massive increase in raw processing power, the April 23 outage suggests that the software orchestration layer responsible for distributing inference tasks across the cluster is struggling with peak-load management. Engineering sources within xAI, speaking on the condition of anonymity, noted that the outage was exacerbated by a scheduled firmware update to the networking fabric that connects the primary GPU clusters in Memphis.
In a brief technical update posted to the official xAI status page, the company confirmed that it was investigating intermittent latency and availability issues affecting the Grok-3 inference engine. By 16:00 UTC, xAI reported that service had been restored to 95% of users, though some continued to experience rate limit warnings as the system cleared the backlog of queued requests. The company has not yet released a full post-mortem report but indicated that additional capacity is being provisioned to mitigate future disruptions.
This event marks the fourth major service disruption for Grok in the first half of 2026. As xAI continues to integrate Grok more deeply into the X platform and its standalone mobile application, the user base has reportedly grown by 40% quarter-over-quarter. The persistent high demand errors underscore the difficulty of maintaining 99.9% uptime while simultaneously deploying cutting-edge large language models that require unprecedented levels of energy and synchronization across distributed hardware.