In a move that has stunned both fans and the tech community, YouTube megastar PewDiePie (Felix Kjellberg) has ventured deep into the world of advanced artificial intelligence development. The creator has unveiled a custom-built AI chat system, nicknamed a “council of AIs,” where multiple independent bots discuss, debate, and vote on the best response before replying to user prompts.
This project marks a fascinating evolution for one of YouTube’s most influential figures. Blending open-source machine learning models, high-end GPU infrastructure, and custom-built software architecture, PewDiePie has created a self-contained AI ecosystem that runs entirely offline, independent of any major cloud provider.
In his video titled “STOP. Using AI Right Now,” PewDiePie detailed how his setup operates. He makes it very clear early on that he is not a fan of how everyone has jumped on the AI bandwagon.
“It’s just become the hot next buzzword and it’s very annoying,” — Felix Kjellberg aka PewDiePie.
“Okay so what AI did I run? First we have to make some things crystal clear. I do not f-with image generation or video generation, ” he goes on to say.
Powered by 10 high-end GPUs, the system runs several large language models (LLMs) ranging from 70 billion to 245 billion parameters, all hosted locally, without reliance on ChatGPT, Gemini, or other third-party APIs.
The architecture follows a multi-agent structure:
Interestingly, emergent behaviors began to appear. Some bots started “colluding” by endorsing one another’s responses. To address this, PewDiePie implemented anti-bias algorithms and reputation scoring, reducing the influence of dominant agents and maintaining fairness within the system.
PewDiePie’s experiment is more than a novelty; it’s a prototype of self-sovereign AI infrastructure. Because the system operates locally, it grants him full control over data, model tuning, and privacy.
This approach aligns with the growing movement toward edge-based intelligence, where AI computation happens on personal or local hardware instead of remote data centers. His 10-GPU workstation mirrors configurations typically used in professional research labs, showcasing how individual creators can now experiment with advanced AI independently.
The YouTuber is also donating idle GPU power from his setup to Folding@home, a distributed computing network supporting medical and scientific research, under the team name “Team Pewds.”
Beyond its technical ingenuity, the concept of a “council of AIs” poses intriguing philosophical questions. How might humans govern artificial intelligence when machines themselves can deliberate and form consensus?
Researchers at Stanford, Anthropic, and OpenAI have explored similar multi-agent systems where AI models exhibit cooperation, negotiation, and dissent.
This project marks PewDiePie’s transition from commentary and gaming into hands-on technological innovation. Over the past year, he has voiced skepticism toward centralized AI platforms, advocating instead for personal, user-owned AI systems. His offline setup reflects both a creative experiment and a philosophical stance against the growing dominance of corporate AI ecosystems.
He plans to refine his models using reinforcement learning, drawing from the council’s voting history to improve decision-making. Future updates may include voice personas, visual interfaces, and even a public demo of the system.
“My next goal is to make them argue less and think more,” he quipped.
PewDiePie’s AI experiment may influence how independent developers approach the next wave of artificial intelligence, one driven by collaboration among multiple AIs rather than single, monolithic systems. Companies like Hugging Face, Anthropic, and OpenAI are already exploring similar architectures under the banner of multi-agent coordination.
You can see his video here: