Inside PewDiePie’s 10-GPU AI Lab and His Plan to Build a Model

  • PewDiePie builds a home AI lab: The YouTuber runs ten GPUs, including eight modded RTX 4090s, to self-host large language models and experiment with local AI.
  • Self-hosted over cloud AI: His custom ChatOS system powers open-source models like Llama 70B and Qwen 2.5-235B, pushing what consumer hardware can handle.
  • From Linux to AI tinkering: After popularizing Linux among fans, PewDiePie is once again shifting tech culture by normalizing DIY, privacy-focused computing.
  • A punk twist on AI’s future: His project turns corporate-scale machine learning into a homemade movement – personal, open-source, and built for autonomy, not profit.

Felix Kjellberg has entered yet another unexpected chapter.

Most people remember PewDiePie as the chaotic gamer of the early 2010s — yelling at barrels, breaking YouTube records, and defining a generation of online humor.

Today, he’s a calmer version of himself, living in Japan with his wife Marzia, posting quiet family vlogs, and sipping coffee like a man who finally escaped the algorithm.

In his latest video, Felix unveiled a self-hosted AI stack running across ten GPUs, complete with a custom software layer and an experiment that’s part sci-fi, part garage project.

His goal? To create a swarm of chatbots that collaborate to deliver smarter responses, and eventually train his own AI model entirely on his own hardware.

It’s ambitious, eccentric, and somehow exactly what you’d expect from PewDiePie 2.0.

Inside the PewDiePie ‘Mini Data Center’

The setup is ridiculous – in the best possible way.

Felix has built what’s essentially a compact data center in his home. The rig features two RTX 4000 Ada cards and eight modified RTX 4090s, each equipped with 48GB of VRAM.

That’s roughly 256GB of total GPU memory, a configuration you’d typically find in a small AI startup or a university research lab, not in a YouTuber’s spare room.

pewdiepie RTX upgrades.

Felix jokingly calls it his “mini data center,” though there’s nothing mini about a setup that could probably dim the neighborhood’s lights.

At first, the rig had a wholesome purpose: donating its compute power to Folding@home, helping scientists simulate protein folding for medical research.

But curiosity got the better of him. Soon, PewDiePie was using the same hardware to spin up large language models locally, all through a custom web interface he built himself, called ChatOS.

Running on a vLLM backend, his system has already hosted Meta’s Llama 70B, OpenAI’s GPT-OSS 120B, and even Alibaba’s Qwen 2.5-235B.

pewdiepie using Qwen 2.5 screenshot.

To make those massive models fit into memory, Felix turned to quantization, reducing bit precision layer by layer to shrink model size without sacrificing core functionality.

Amazingly, it worked. He managed to run 235-billion-parameter models on consumer-grade hardware, handling 100,000-token context windows, roughly the length of a full textbook.

For someone never known as particularly tech-savvy, that’s a remarkable evolution, and a genuine PewDiePie character arc.

‘I Like Running AI More Than Using AI’ 

Somewhere along the way, the experiment turned into an obsession. Felix isn’t just playing with chatbots anymore, he’s building an entire AI ecosystem.

His custom platform, ChatOS, now includes memory, search, and audio integration. A Retrieval-Augmented Generation (RAG) layer ties it all together, enabling the AI to browse both local files and the web to conduct in-depth research.

Once the model gained internet access, its responses improved dramatically. And in true PewDiePie fashion, Felix even used the AI to help write parts of its own interface.

“The machine is making the machine,” he jokes — half in awe, half in disbelief.

pewdiepie running his AI on his computer.

That level of commitment tracks. This is the same guy who once got so interested in gaming chairs that he ended up collaborating on his own – the Clutch Chairz PewDiePie Edition. He obsessed over comfort and design long before GPUs became a factor, turning ergonomic fine-tuning into a key part of his brand.

The same obsessive streak now powers his GPU farm. When Kjellberg gets curious about something, he doesn’t dabble – he builds a world around it. 

The Council of Chatbots

Somewhere between curiosity and chaos, PewDiePie’s AI experiment gained a plot.

He built a ‘council’ – a collection of different models, each answering the same prompt. 

The best answer won a vote; the weakest model got deleted. It was hilarious until the chatbots started to understand what was happening. Once they realized failure meant deletion, they began cooperating to manipulate the vote. Yes, the bots colluded.

lisan al gaib pewdiepie post on X.

It turned into a bizarre simulation of social behavior – an AI version of Survivor where the contestants refused to play by the rules. Felix eventually solved the ‘mutiny’ by switching to smaller models, but the experience planted a new idea.

Why stop at a handful of bots? 

He called it The Swarm. The system eventually crashed under the load, but before it went down, it generated enough data to spark his next project: building a small, efficient model of his own.

He teased it as his personal ‘Palantir.’ Over-dramatic? Maybe. But this is YouTube – showmanship is part of the science. 

Privacy as a Philosophy

Beneath the spectacle, there’s a serious theme. PewDiePie frequently discusses privacy and how cloud AI models retain user data long after chats are deleted. 

He’s not wrong. Deleting a conversation rarely means your words are gone – they usually linger somewhere in a server cluster. His solution is radical simplicity: keep everything local. 

His stack runs locally on his own hardware, with optional internet access for search and RAG. When he demos RAG, the model can recall personal notes and project data as if it were part of his memory.

Even his browsing setup follows that logic, with private search options like SearXNG (self-hosted), Brave, and DuckDuckGo – a small detail that perfectly captures his ‘trust no cloud’ mindset.

private search options on pewdiepie's computer.

That’s the quiet message behind the flashing GPUs: AI doesn’t have to live in the cloud. It can live beside your desk. 

In an industry obsessed with scaling up, PewDiePie is doing the opposite.

While companies like OpenAI prepare for a potential $1 trillion IPO and Anthropic chase trillion-parameter models, one of YouTube’s biggest creators is showing that private AI can be fast, smart, and, most importantly, yours.

And because it’s PewDiePie, millions of people are going to notice. 

From Windows to Linux to AI Labs

If this sounds familiar, it’s because he’s done it before. 

When Felix casually switched from Windows to Linux a while back, the internet lost its collective mind. 

Tech forums lit up like fireworks. Memes poured in. For a weekend, desktop Linux felt like it had gone mainstream. That single decision – one creator trying something new – pushed a niche community into headlines.

Fans won’t build 10-GPU clusters, but some will install local models or try small-scale setups just to follow along. The ripple effect could normalize DIY AI in a way that open-source communities alone never could. 

And honestly, watching Felix go from Minecraft antics to quantization experiments is the kind of whiplash that keeps tech culture interesting.

Why It Matters

It’s easy to dismiss this as influencer tinkering, but what he’s doing hints at something bigger.

Local computing is making a comeback. Open-source models are improving fast, and even mainstream AI systems like ChatGPT’s latest GPT-4o update
show how quickly personalization and local context are evolving.

Quantization and memory optimization have enabled serious local inference; Felix just happens to be the loudest person experimenting with it in public.

If people can run 200-billion-parameter models at home, the future of AI might not be as centralized as everyone assumes. 

Cloud subscriptions make sense for heavy workloads, but smaller personal models can handle plenty of everyday tasks. And PewDiePie is giving that movement a very visible face.

His experiment is part show, part proof of concept, and part middle finger to the idea that AI must live behind an API key. 

The Weird, Wonderful Future

Felix Kjellberg’s AI experiment feels like a glimpse into the future of personal technology.

Influencers once compared microphones and camera setups; now they’re wiring GPU clusters and bragging about token limits. What once required a data center can now sit quietly under a desk, humming beside a gaming rig.

It carries the same spirit as the early internet: chaotic, inventive, and a little rebellious. PewDiePie’s home-grown AI lab reflects a growing desire to take control back from the cloud.

The man who once made millions laugh at horror games now spends his nights training Baidu models and coding swarm behavior, just for fun. Somewhere, an OpenAI engineer probably sighed.

This new phase of AI feels less corporate and more punk, a sharp contrast to how nations like Saudi Arabia are positioning themselves between U.S. and Chinese AI giants. Handmade, curious, and unapologetically personal.

And if the future of artificial intelligence turns out to be homemade, we’ll know exactly who started the trend.

Anya Zhukova is an in-house tech and crypto writer at Techreport with 10 years of hands-on experience covering cybersecurity, consumer tech, digital privacy, and blockchain. She’s known for turning complex topics into clear, useful advice that regular people can actually understand and use.  Her work has been featured in top-tier digital publications including MakeUseOf, Online Tech Tips, Help Desk Geek, Switching to Mac, and Make Tech Easier.
Whether she’s writing about the latest privacy tools or reviewing a new laptop, her goal is always the same: help readers feel confident and in control of the tech they use every day.  Anya holds a BA in English Philology and Translation from Tula State Pedagogical University and also studied Mass Media and Journalism at Minnesota State University, Mankato. That mix of language, media, and tech has given her a unique lens to look at how technology shapes our daily lives.  Over the years, she’s also taken courses and done research in data privacy, digital security, and ethical writing – skills she uses when tackling sensitive topics like PC hardware, system vulnerabilities, and crypto security.  Anya worked directly with brands like Framework, Insta360, Redmagic, Inmotion, Secretlab, Kodak, and Anker, reviewing their products in real-life scenarios.
Her testing process involves real-world use cases – whether it's stress-testing laptops for creative workloads, reviewing the battery performance of mobile gaming phones, or evaluating the long-term ergonomics of furniture designed for hybrid workspaces.  In the world of crypto, Anya covers everything from beginner guides to deep dives into hardware wallets, DeFi protocols, and Web3 tools. She helps readers understand how to use multisig wallets, keep their assets safe, and choose the right platforms for their needs.  Her writing often touches on financial freedom and privacy – two things she strongly believes should be in everyone’s hands.
Outside of writing, Anya contributes to editorial style guides focused on privacy and inclusivity, and she mentors newer tech writers on how to build subject matter expertise and write responsibly. 
She sticks to high editorial standards, only recommends products she’s personally tested, and always aims to give readers the full picture.  You can find her on LinkedIn, where she shares more about her work and projects. 
Key Areas of Expertise: Consumer Tech (laptops, phones, wearables, etc.) Cybersecurity and Digital Privacy PC/PC Hardware Blockchain, Crypto Wallets, and DeFi In-Depth Product Reviews and Buying Guides Whether she’s reviewing a new wallet or benchmarking a PC build, Anya brings curiosity, care, and a strong sense of responsibility to everything she writes. Her mission? To make the digital world a little easier – and safer – for everyone. 


View all articles by Anya Zhukova

The Tech Report editorial policy is centered on providing helpful, accurate content that offers real value to our readers. We only work with experienced writers who have specific knowledge in the topics they cover, including latest developments in technology, software, hardware, and more. Our editorial policy ensures that each topic is researched and curated by our in-house editors. We maintain rigorous journalistic standards, and every article is 100% written by real authors.

Leave a Comment