Gemini prompts stalled globally as Google Cloud issue disrupted services
Users and enterprises reported hung Gemini prompts and slow responses; trackers showed outage peaks at 2:30 p.m. EDT and residual reports into the evening.

Prompts to Google’s generative-AI assistant Gemini stalled for users worldwide on Feb. 21, 2026, producing hung processing loops, missing chat history and slow or failed responses in both the consumer app and Workspace integrations. Public monitors reported a surge of complaints that peaked around 2:30 p.m. EDT and continued at lower levels into the early evening.
SiliconANGLE’s reporting traced the disruption to an identity and access management misconfiguration inside Google Cloud that, it said, cascaded into outages or intermittent failures for core cloud services including App Engine, Firestore, Cloud SQL, BigQuery and Memorystore. Cloudflare, which reported its own customer-impacting problems, told CRN and was quoted in SiliconANGLE: “This is a Google Cloud outage.” The company said specific features such as Workers KV, Access authentication, Workers AI, Stream and parts of its dashboard were affected because of dependency on Google Cloud.
The outage was visible across downstream services. SiliconANGLE and social media posts cited intermittent failures in Gmail, Drive, Docs, Calendar, Meet and Chat; users also reported nonresponsive voice search and assistant features and connectivity problems for Google Home and Nest devices. Public trackers such as Downforeveryone and similar sites recorded recent reports from countries including Indonesia, India, Hong Kong, France, Malaysia, the United States, Trinidad and Tobago, Morocco and Portugal.
Google’s own status dashboards have logged multiple Gemini-related incidents over the past two years, with entries noting symptoms and engineering investigations. Past entries include an October 2024 problem affecting Workspace extensions that Google said was “limited to only enterprise users,” and mid-2025 incidents that lasted from under an hour to several hours. Status messages during those incidents repeatedly warned that “workaround: none at this time” and advised that engineering teams were investigating. One Google status notice included the user-facing line: “Gemini is currently on a break. Give it a moment and try again later.”

Independent diagnostic guidance from SiliconANGLE contributors underscored two practical checks for users: verify whether a widespread outage is being reported across Google’s fragmented health dashboards and test client-side issues such as browser cache or extensions. As John Furrier wrote in diagnostic commentary, “The first step in troubleshooting is to check for a widespread outage. This can be challenging because Google's service health information is fragmented across multiple dashboards.” He added that “The generic 'technical error' message encountered when Gemini's image generation fails is not a simple bug but a symptom of a complex, multi-layered system,” and warned that lack of transparency leaves individual users to troubleshoot without clear direction.
The outage highlights broader risks as enterprises and public services increasingly embed large-language models and cloud-hosted AI into workflows. Health-care providers, community service organizations and smaller businesses that rely on Workspace tools and AI integrations for scheduling, documentation or patient-facing triage could face disruptions if cloud dependencies cascade. That vulnerability raises equity concerns: communities with fewer resources to provision redundant systems or local backups may experience disproportionate disruptions to essential services when large cloud providers falter.
Google, Cloudflare and other affected vendors did not publish a joint root-cause postmortem in the material available to reporters; SiliconANGLE’s technical attribution and Cloudflare’s statement remain the strongest indications of cause in public sources. For users experiencing problems, sources recommend checking multiple Google status dashboards for model- or product-specific notices, trying Incognito mode and clearing browser cache and cookies as interim troubleshooting steps while awaiting an official engineering report.
Know something we missed? Have a correction or additional information?
Submit a Tip

