Four major AI platforms hit overlapping trouble today.

ChatGPT, Claude, and Grok confirmed outages. Gemini showed a spike in user reports. That overlap is rare. It is also a design review.

We still treat frontier AI as electricity: always on, always someone else’s problem. Then a few APIs and a few vendors blink together, and knowledge work stops.

Fragile systems concentrate. One provider. One region. One model. One workflow that cannot run if the API is dark.

Antifragile systems keep optionality. Local models for the work that cannot wait. A second provider already wired in. Offline drafts, cached context, and a human fallback that does not require a status page. The goal is not to abandon the cloud. It is to stop building companies that freeze when the cloud blinks.

Nvidia’s deal for Hugging Face — about $13 billion — is the same point from the other side. Open weights and portable models are not a hobby. They are a hedge against concentrated downtime. They are also a reminder: the hedge only works if you can run the model when the vendor cannot.

If your operating model dies when the logos go gray, you do not have an AI strategy. You have a single point of failure with a chat interface.

When that happened this morning, which of your workflows actually stopped — and which already had a second path?