Product1 distinct publisher3 min readPublished
Grok went down for three and a half hours on Thursday and Claude's error window opened the same minute, but no vendor will say whether the two shared a building, which leaves customers explaining an outage they cannot trace.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
The person who had to explain Thursday to their own customers had two status pages and a clock. SpaceXAI logged Grok's incident as a "models outage" starting 6:30AM PT and running three and a half hours [3], which puts recovery near 10:00AM PT [1]. Anthropic's page showed elevated errors for Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5 opening at roughly the same minute and marked resolved at 9:16AM PT [6], a window of about two hours and 46 minutes [2]. Neither page said whose hardware sat underneath.
The only party that put the two things in the same sentence was SpaceXAI, which apologized to Grok users and then to "impacted compute partners" it did not name [1]. It has not said what failed in Memphis, and Elon Musk said the company was taking corrective action so it would not happen again [4]. Anthropic leases compute from Musk's AI firm under a deal signed earlier this year [7], and has not said whether Thursday's errors had anything to do with the Memphis site [8]. Anthropic and OpenAI did not respond to Engadget's requests for comment [10].
OpenAI's own incident, elevated errors across ChatGPT and Codex, was reported from around 7:30AM PT and cleared at 9:55AM PT [9], about two hours and 25 minutes [3] that started an hour later than the other two [4]. Nothing any of the three companies has said places OpenAI's outage in Memphis [5][9]. Engadget's framing is that the timing "roughly lines up" [5], and that is where the evidence stops: one supplier has confirmed that a single facility can take out customers other than itself, and has kept the roster private.
Teams that route across two model providers usually describe that to themselves as redundancy. What the Memphis apology shows is that redundancy bought at the vendor layer can be cancelled at the facility layer, and that the customer holds no document that would tell them which of their two providers sits in the same hall as the other. Grok's own outage was the longest of the three windows, about 44 minutes longer than Claude's [5], which is the shape you would expect if the site operator is also the site's largest tenant, though nobody has published a cause to confirm it [4].
The useful exercise here is two columns rather than a diagram. For every model dependency you run in production, write the vendor you pay in column one and the operator of the physical capacity in column two. Then sort each row: some pairs are known to sit in different buildings, some are known to share one, and the rest are simply unknown. Anthropic's lease from SpaceXAI [7] is the unusual case where a buyer can fill in column two from public reporting. Most rows will read unknown, and Thursday is the evidence that treating unknown as different was an assumption, not a fact. A fallback you cannot trace to a different building only protects you against your vendor's software; it tells you nothing about your vendor's landlord.
Ranked by verification strength, evidence, and original report placement.
SpaceXAI posted an apology saying: "We are sorry for the issues you may have experienced with Grok following an outage at our Memphis compute center this morning. We'd also like to apologize to our impacted compute partners." The affected partners were not named.
The same SpaceXAI post, dated September 3, 2026, said "All systems have now been restored and are functioning nominally."
According to SpaceXAI's status page, Grok was down for three and a half hours beginning at 6:30AM PT on Thursday; the incident was labeled "models outage" and apparently affected Grok on X as well as its Android and iOS apps.
SpaceXAI did not disclose what caused the issue in Memphis; Elon Musk said in a post that the company was "taking corrective action to ensure this does not happen again."
Engadget reported that it is not entirely clear which of SpaceXAI's "compute partners" may have been affected, but that the timing of the incident roughly lines up with issues experienced by other AI platforms Thursday morning.
Anthropic's status page cited "elevated errors for multiple models" starting around 6:30AM PT Thursday, affecting Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5, and marked the issue resolved by 9:16AM PT.
Distinct publishers with included, body-backed reporting in this cluster.
2 articles · September 3, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
invest
Anthropic Cuts Cache-Read Prices by 75%; Cache Reads Were ~60% of a Heavy Agent's Bill Before the Cut2 distinct publishers
build
Developer habit, priced at $965B: what Anthropic's run actually proves1 distinct publisher
build
OpenAI's Cursor cutoff reclassifies model access as a supply-chain dependency8 distinct publishers
product
OpenAI's roadmap now fits inside one president's span of control1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Primary documents, single messenger
Every time in this story — 6:30AM PT, 9:16, 9:55, three and a half hours — comes off a status page the vendor can edit, plus a post from Musk, relayed by one newsroom. Those are primary artifacts and Engadget quotes them tightly, which is why this scores respectably. But the single fact that would turn a coincidence into a shared-infrastructure story, namely that Claude's errors ran on Memphis hardware, is precisely the fact nobody put on the record, and the second copy of the piece in our coverage adds a URL, not a source.
Three services down, blast radius unmapped
The real-world footprint is not in doubt: Grok on X and both mobile apps, three named Claude versions, ChatGPT and Codex, all inside one morning, with Downdetector traffic as the one non-vendor witness. What cannot be sized is how much of that traced back to Memphis. The compute lease is the only disclosed pipe between SpaceXAI and Anthropic, and no party quantified capacity, customers or requests — so the incidents are countable while their common ground is not.
Hedged words, suggestive arrangement
Engadget's sentences are careful — "apparently," "not entirely clear," "roughly lines up" — but the layout does work the wording refuses, placing OpenAI's outage directly beneath a Memphis apology it has no disclosed connection to. OpenAI's onset was an hour later and its provider relationships never come up, so a reader can leave with a three-lab common-cause story that the reporting itself declines to make. Small overshoot, and it lives in the juxtaposition rather than in any claim.
An apology with no address on it
Consider who chose each word here. SpaceXAI apologizes to "compute partners" without naming one, which discharges the courtesy while protecting the customer list and any contractual exposure. Anthropic, which buys capacity from the firm that just went dark, says nothing about a link and does not return a reporter's call. Musk promises corrective action while withholding the fault. Not one party with a commercial reason to speak plainly did, and the silence is symmetrical enough to look negotiated.
Firm on the clock, open on the cause
We can state with confidence what broke, when, and who declined to explain it — those facts rest on published vendor records and simple arithmetic. We cannot state that Memphis took Claude down, and one outlet working from status pages was never going to settle that. Confidence sits mid-range for the honest reason: the descriptive layer is solid, the attributive layer is empty, and a single postmortem or one on-record Anthropic engineer would move this sharply in either direction.