# Emad Mostaque Says AI and Internet Infrastructure Are Headed for a Harder, Faster Future
Emad Mostaque used a talk at the TechBBQ startup conference in Copenhagen to argue that artificial intelligence is entering a phase where technical capability, persuasion and infrastructure fragility are converging. The supplied report, dated 2026-09-06, describes a long conversation in which the Stability AI founder and current Intelligent Internet chief sketched a future that is at once more powerful and more brittle.
Mostaque’s central claim is that the internet and the systems built on top of it are far less stable than many assume. He says the underlying infrastructure is held together by twigs and predicts that the internet may go offline in the next few years. That prediction is not presented as a certainty but as a warning about how exposed global systems have become to attack and failure.
The talk also framed AI as a geopolitical and security issue. Mostaque said systems beyond public understanding are already circulating in Washington, and he described a world in which frontier models are able to hack, persuade and coordinate at a level that changes both offense and defense. He argued that governments and militaries will increasingly divert resources toward AI-powered systems, drones and robots because the cost of such capability is falling rapidly.
A major theme was sovereignty. In Mostaque’s framing, sovereignty means the ability to resist power exerted over you, and that in the AI era it depends on owning the stack rather than renting it. He argued that countries and individuals need control over their own models, context and objectives because otherwise they end up importing someone else’s morals, ethics and incentives.
The report says he linked that concern to persuasive capability. If AI systems can outperform humans in debate, analyze data, and carry out tasks with little supervision, they become more than assistants. They become instruments of influence. Mostaque warned that such systems could weaken critical thinking, tilt politics and eventually shape national decision-making in ways that make human oversight less meaningful.
The economic argument was just as stark. Mostaque suggested that competent human labor could lose value quickly once AI systems become sufficiently capable. He pointed to models that can already do math without obvious mistakes, handle tax returns and spin up large numbers of agents. In that environment, he said, companies may be able to replicate whole workforces as virtual agents that cost very little to run.
The report also highlights his interest in speed and hardware. He cited chip-level model approaches, open reasoning datasets and open-weight models as signs that the industry is moving quickly toward a world where standards must be defined around these systems. His message to the audience was not that AI is arriving someday, but that the practical consequences are arriving now.
Whatever one makes of the forecast, the talk captures the current mood around frontier AI: less celebration of raw capability, more concern about concentration, dependence and control. Mostaque’s argument is that the technical race is already remaking politics and labor faster than public institutions can adapt.
