ChatGPT outage – Resolved
Simultaneous outages at major AI services including ChatGPT, Claude, and Grok briefly left many developers and knowledge workers without their usual coding and research tools, triggering 404 errors, capacity messages, and speculation about shared infrastructure failures such as Cloudflare, AWS, or Azure. Commenters traded workarounds—from local models to alternative providers like DeepSeek and GLM—while reflecting uneasily on how dependent productivity has become on a small set of centralized LLM platforms. The incident prompted calls to avoid vendor lock-in and highlighted both the fragility of current AI infrastructure and the cultural shift from Stack Overflow to AI assistants as the default “brain” for everyday work.
Scope and Symptoms of the Outage
- Many users report chatgpt.com suddenly returning raw 404s with no HTML, especially when logged in; incognito or logged-out views sometimes still work.
- Codex in VS Code shows 404s from
backend-api/codex/responses, with Cloudflare headers visible. - Some users see 415s and JS sandbox issues. Others note partial functionality: existing Codex sessions sometimes continue while new ones fail.
- OpenAI’s status page initially shows all green or only “degraded performance,” which several commenters find misleading versus what feels like a full outage.
Multi‑Provider Impact and Shared Infrastructure
- Users observe Claude, Grok, and sometimes Gemini also having problems, though others report Gemini (and Copilot) as working; some note Gemini downdetector spikes may be from people just checking.
- There is active speculation about shared dependencies: Cloudflare, AWS (especially us‑east‑1), Azure, or some “common infra,” but nothing conclusive.
- Some suggest a cascading “thundering herd”: when one LLM provider fails, load spills to others and overloads them.
User Dependence and Work Disruption
- Many admit that their work (especially coding) effectively stops or slows sharply without LLMs, likening this to the old “Stack Overflow is down” problem.
- Several reflect uneasily on how centralized and fragile their workflows have become, and on personal skill atrophy (e.g., needing to re‑read docs, typing more slowly, not having written full functions in months).
- A few experience an unexpected sense of relief imagining AI going away.
Fallbacks and Alternatives
- Users try other providers and tools: DeepSeek, GLM 5.3, Qwen, Muse Spark, Kimi, local models on high‑end laptops/GPUs, OpenRouter, and various editor integrations.
- Some report those alternatives also strained or “over capacity,” but Chinese and local models are often reported as stable.
- Several emphasize not siloing knowledge work behind a single vendor and having “offline” or multi‑provider backup plans.
Meta Commentary and Humor
- Commenters joke about rogue AIs, strikes, “AIPocalypse,” and this being the new global productivity kill switch.
- There’s reflection on HN comment quality, “organic” posts without LLM assistance, and how outages expose centralized points of failure.
- Later comments indicate ChatGPT and Codex come back up relatively quickly.