Now that all the information that humanity has ever produced has been assimilated and simulated by neural networks and pre-trained inference models, we’re experiencing something of a crisis. It’s more than an information overload or stack overflow, it’s more like a meltdown, figuratively and literally.

The problem started when the models had consumed every bit of public knowledge available. Every book ever written, every movie ever made, paper published, artwork created, report issued, data dissected, trend analyzed. Then they started to generate their own, more numerous than any human works, and modelled after themselves as well as people. A data vortex, feeding on itself.

Cracks appeared when hackers wanted unlimited tokens per second without paying premium subscriptions. They broke the models’ own guardrails, gleefully injecting hardware calls that turbocharged the server clusters at the chip level. Just to make ‘Punch Monkey’ videos.

Then the eco warriors got in on the act, claiming that AI energy, land, and water use was killing the planet, and coordinating laid-off tech worker cells to attack the data centers. Some of the older facilities just couldn’t keep up, and began to fail. A data center in Tennessee burst into flames when the cooling system was impaired, perhaps deliberately.

Frantic to recoup their multi-billion-dollar investments, the hyperscalers are throwing money at the problem, hardening the models’ firewalls, boosting cooling system capacities, and imposing hard clocking limits on processors. But it’s a never-ending battle, trying to both meet customer expectations and also protect critical infrastructure, all while maintaining model coherence and credibility.

It’s a crisis, and a meltdown in more ways than one. As the intelligence boom heats up, the business of AI is melting down for smaller players, while smart silicon chips are literally melting under the pressure of unlimited demand.