Cardano (ADA) Founder Charles Hoskinson Floats Nation-State Theory After 90-Minute AI Outage
Cardano founder Charles Hoskinson suspects a nation-state attack after ChatGPT, Claude and Grok failed within 90 minutes on Thursday.
AI SummaryAI
- ChatGPT outage reports peaked at roughly 38,000 on Downdetector, with 80% flagging ChatGPT.
- Charles Hoskinson claimed a nation state likely brought down Claude, ChatGPT and Grok simultaneously.
- Anthropic draws over 300 megawatts across 220,000 Nvidia chips at Memphis Colossus 1.
- OpenAI applied a mitigation at 11:17 a.m. EST and resolved its incident by 12:55 p.m.
Three AI Labs Went Dark in 90 Minutes
ChatGPT, Claude and Grok all stopped responding for large numbers of users on Thursday morning, a rare simultaneous disruption across three rival AI platforms. Complaint data on Downdetector shows OpenAI absorbing the heaviest load, with reports peaking at roughly 38,000 — among users who identified a specific product, 80% flagged ChatGPT, 8% Codex and 7% the app. OpenAI's status log shows engineers began investigating elevated errors across ChatGPT and its Codex coding agent at 10:43 a.m. EST, applied a mitigation at 11:17 a.m. and closed the incident by 12:55 p.m. Anthropic's timeline was busier: elevated error rates for Claude Sonnet 5 surfaced at 8:37 a.m. EST and cleared 19 minutes later, before a second incident spanning several Claude models ran from 9:26 a.m. until 12:16 p.m. EST. Elon Musk's xAI confirmed an outage touching Grok's web and mobile apps, its presence on X and two U.S. API regions, restoring service by 1:07 p.m. EST. The disruption landed hard on users who lean on these tools daily — chemist Simon Maechling asked on X how anyone could get work done, while Every CEO Dan Shipper joked that ChatGPT's downtime made him “immediately 50 IQ points dumber.”
Hoskinson Points to Shared Nvidia Dependency
Cardano founder Charles Hoskinson, who also co-founded Ethereum, offered a more provocative reading of events: it “looks like a national state” brought down Claude, ChatGPT and Grok at once, he wrote on X — implying a government actor rather than a hacker or a routine technical fault. His reasoning rests on a shared supply chain: the three models all run on Nvidia chips, he argued, whereas Google's Gemini was the lone major assistant spared, since it serves on its own TPUs. The timeline gives the theory its texture — Grok and Claude users began reporting errors near 9 a.m. Eastern, and ChatGPT complaints jumped from 5,000 to more than 22,000 within 10 minutes roughly 90 minutes later. The damage was lopsided: Claude and Grok each peaked near 1,500 reports while ChatGPT cleared 35,000. Notably, no cloud provider declared a fault, xAI never logged an incident at all, and Anthropic users instead hit capacity walls citing “unexpected capacity constraints.” OpenAI listed 19 affected components, Anthropic blamed its Opus models, and no lab has reported an attack. In our read, the shared-chip argument is suggestive but not dispositive — Nvidia hardware in separate buildings does not normally fail in unison.
he wrote on Xhttps://x.com/IOHK_Charles/status/2095532822933246208?s=20
Memphis Colossus Draws Scrutiny
Attention quickly turned to a single building in Memphis. SpaceX merged with xAI in February, then rented Anthropic the full compute capacity of Colossus 1, the cluster built for Grok; Anthropic draws more than 300 megawatts across 220,000 Nvidia chips there, just under half of xAI's roughly 500,000-GPU fleet. Two of Thursday's casualties therefore sit on the same racks, though neither company has linked the outage to that co-location. The episode also lands amid a broader run of infrastructure fragility: a Cloudflare outage hit crypto platforms last November, two AWS failures knocked services offline weeks before that, and Anthropic went dark alone in a similar incident in March. The failures have energized the decentralized-AI camp, too. Analyst Wendy O tied the disruption to the case for decentralized AI systems, name-checking Midnight — the privacy sidechain from Hoskinson's Cardano ecosystem, a design philosophically closer to the Zcash (ZEC) privacy model than to a hard fork of an existing chain. The pitch sits within the wider DeFAI thesis that AI-powered decentralized finance can strip single points of failure out of both compute and settlement layers. Readers tracking the market in real time can follow live spot and futures prices on MEXC.
Decentralization Pitch Meets Concentration Reality
COINOTAG's take: whatever the root cause proves to be, Thursday was a stress test of concentration, and the blast radius is measurable. The most load-bearing primary document in this story is the official SEC filing from SpaceX and xAI, which states that Grok served approximately 117 million monthly active users as of March 2026 — a user base that large, dependent on one Memphis building and one chip supplier, is precisely the fragility that Bitcoin's distributed validator architecture was designed to avoid. Whether the failures were accidental or adversarial remains undisclosed; until a confirmed post-mortem lands, Hoskinson's nation-state framing should be treated as a hypothesis, not a finding. What it has already done is hand the decentralized-AI narrative its most vivid case study in months.
Related Tags

AI-generated, AI-reviewed, under COINOTAG editorial oversight.


