Three AI outages, one lesson about vendor risk

Does yesterday's triple AI outage change the risk profile of AI?

Outages happen. Three at once, ChatGPT, Claude and Grok, that is new. Most outages resolve in hours, and this one did too. But three unrelated pieces of software, three separate infrastructures, going down at the same time points at something underneath all of them. This was not an isolated mistake. I ran into another bug today making the image for this post, a first for me in four years of very intensive GPT and Claude use.

So yes, this does change the risk profile. Not because AI got riskier overnight, but because the odds that you can rely on any one provider's independence just moved, in public, all at once. It is still vendor and technology risk, not a new category. It is the same risk moving from theoretical to observed.

And that piece of risk is not going away. AI is a very complex and potentially vulnerable system using a supply chain with real bottlenecks. Add geopolitical, energy and climate chaos, and this risk will not decrease.

There is also a shift worth naming. Traditional supply lines have redundancy, stock, wait times, moving goods around, and users expect to wait a bit for delivery. With AI both have changed: little redundancy behind the scenes, and users trained to expect immediate, every time, with no tolerance for a blank screen. The AI majors have become the face of an enormously complex system, carrying an expectation that system was perhaps promising but cannot offer.

Of course, stay calm. Build for the occasional outage, not around the fiction that it will not happen. And keep in mind that building your own AI infrastructure, keeping your own data and not betting your company on AI may make a bit more sense than it did before yesterday.