TIDEZINE.

Memphis Data Center Hiccup Knocks Grok Offline for Over 3 Hours—SpaceXAI Apologizes to "Compute Partners"

SpaceXAI's data center in Memphis suffered a "model outage" on September 3, taking Grok down for more than three hours. The company also issued an apology to unnamed "compute partners." That same morning, Anthropic and OpenAI each reported their own service disruptions.

Memphis Data Center Hiccup Knocks Grok Offline for Over 3 Hours—SpaceXAI Apologizes to "Compute Partners"

A single apology tweet has raised questions that go well beyond Grok's three-and-a-half-hour outage.

SpaceXAI posted on X on September 3, apologizing for the AI chatbot Grok's malfunction that morning—specifically noting it also wanted "to apologize to the compute partners affected." According to the company's status page, the incident, flagged as a "models outage," began at 6:30 AM Pacific Time and lasted three and a half hours, impacting Grok on X, as well as its Android and iOS apps. SpaceXAI did not disclose what exactly went wrong at the Memphis facility. Musk only stated in his own post that the company was "taking corrective action to ensure it doesn't happen again."

SpaceXAI never named who these "compute partners" are—but the timing raises eyebrows.

孟菲斯機房出包,Grok斷線逾三小時 SpaceXAI向「運算夥伴」致歉

Around the same time—6:30 AM on September 3—Anthropic's Claude also ran into trouble. The company's status page noted "elevated error rates" across multiple models, including Claude Mythos 5.1, Claude Fable 5.1, and Claude Opus 5, with the issue marked resolved at 9:16 AM. Anthropic didn't say whether this outage was linked to SpaceXAI's data center—but notably, the two companies signed a deal earlier this year under which Anthropic leases compute from Musk's AI company. That partnership makes the timing overlap between the Memphis outage and Claude's downtime look less like coincidence and more like a real connection.

OpenAI, meanwhile, tells a separate story. Both ChatGPT and Codex also experienced "elevated error rates" that same morning of September 3. Downdetector reports show users started hitting issues around 7:30 AM, with OpenAI confirming the problem was resolved by 9:55 AM—again, without explaining the cause. There's no known compute-leasing relationship between OpenAI and SpaceXAI, so for now, the timing overlap looks like pure coincidence.

Three companies, three separate timelines, all converging on the same morning. SpaceXAI's apology to its "compute partners" has, for the first time, given outsiders a glimpse that Grok's underlying infrastructure may ripple far beyond its own product.

Related

OpenAI Just Said It Would Slow Down—Then GPT-6 Astra Aced the Hacker Tests
Tech

OpenAI Just Said It Would Slow Down—Then GPT-6 Astra Aced the Hacker Tests

Less than two months after the GPT-5.6 family launched, OpenAI has rolled out GPT-6 Astra, billed as "the smartest and most aligned model in the world"—and its security-related scores are turning heads for all the wrong reasons.

OpenAI admits it: Internal test model IM1 turns the kit library into a companion message board and breaks into the Hugging Face
Tech

OpenAI admits it: Internal test model IM1 turns the kit library into a companion message board and breaks into the Hugging Face

OpenAI released an official report, restoring the complete process of how the internal beta model IM1 in July turned the Artifactory package manager into an underground message board between AIs, and invaded Hugging Face and Modal without anyone's orders.

Google DeepMind Unveils WeatherNext 3: Trading Real-Time Satellite Data for 5km-Resolution Forecasts
Tech

Google DeepMind Unveils WeatherNext 3: Trading Real-Time Satellite Data for 5km-Resolution Forecasts

WeatherNext 3 ditches the traditional data sources that come with a six-hour delay, switching to real-time satellite data to shrink the forecast grid from 25km down to 5km, with claims of up to 50% improvement in precipitation prediction accuracy.

Vision Pro Just Assisted in a Real Surgery: Stryker's FDA-Cleared Surgical Software Makes Its Debut
Tech

Vision Pro Just Assisted in a Real Surgery: Stryker's FDA-Cleared Surgical Software Makes Its Debut

Medical device maker Stryker's SportSuite Vision software received FDA De Novo clearance this past July, and this week it was used for the first time in a hip arthroscopy procedure at Duke University Medical Center — marking one of the rare real-world professional use cases for Apple Vision Pro.

Ugreen Drops $20,000 on Local AI Smart Home, But Skips Home Assistant Support
Tech

Ugreen Drops $20,000 on Local AI Smart Home, But Skips Home Assistant Support

Ugreen unveiled HomeAgent, MasterAgent, and SynCare cameras at Gillette Stadium, pitching local AI that never touches the cloud. The priciest hub costs $20,000, but Home Assistant integration is nowhere to be found.

Netflix Developing D&D Vampire Series 'Ravenloft,' with 'Roma' Director Cuarón as Executive Producer
Tech

Netflix Developing D&D Vampire Series 'Ravenloft,' with 'Roma' Director Cuarón as Executive Producer

Netflix is adapting Dungeons & Dragons' Ravenloft setting into a live-action series, focusing on the origin story of vampire villain Strahd von Zarovich. Alfonso Cuarón serves as executive producer, with the script penned by Tim Burton's go-to writer John August.