Nye's Digital Lab is a weekly scribble about creativity at the intersection of AI and distributed systems.

This week I'm opening all the weights. Open Source software (and now AI) could create a massive alternative network to the proprietary systems.

I believe that open source software has hidden magic.

You can git pull billion-dollar technologies, modify them however you want, and build entire new things on top of them. For free. It's "magic community lego," and I believe it will have a profound influence on the AI and distributed network landscape.

But first let's recap:

DeepSeek, a Chinese company, "blew the barn door off" the most expensive AI systems in the world, using open source approaches that cost a fraction of what tech giants spend. They aren't the only Chinese AI company either. Alibaba's Qwen, Tencent's Hunyuan models, and Baidu's ERNIE (which comes from a company that historically has driven a proprietary strategy) are all embracing openness. Between 2022 and 2023, the number of globally available open-source projects saw an average growth of 29%.(1)

Chinese AI companies are strategically embracing open-source AI to cultivate independent AI capabilities and reduce their reliance on Western technologies.(2) This "scorched earth" is also the strategy of Meta and their Lllama model. This approach isn't merely about technological parity but about fostering a self-sufficient and innovative AI ecosystem within China, the US and around the world.

This isn't another tech story.

It's a preview of a fundamental shift in how technology gets built, who controls it, and what that means for the rest of us.

While companies like OpenAI and Google pour billions into proprietary systems, DeepSeek achieved comparable performance using open source methods and dramatically less compute power. It's like watching a garage band outperform a major label's million-dollar production.

For years I discounted Blender, the open source 3D animation software that's now outperforming expensive tools like Autodesk Maya. Python became an industry standard programming language not because a corporation pushed it, but because developers collectively made it better than any proprietary alternative. The open source game engine Godot is rapidly gaining ground on Unity and Unreal Engine, despite having a fraction of their corporate backing. Nvidia just dropped Isaac Sim, their mega robot training virtual sim. (4) Again, open source.

The internet stack, made of countless libraries of JavaScript, HTML, and various protocols, is entirely open source. Ninety-nine percent of the world's servers run Linux, and GitHub hosts an eternal playground of frameworks, libraries, tools, and projects with millions of contributors. (5) The same pattern emerges with Bitcoin, Layer 2's on Ethereum, and Solana, all blockchain ecosystems built and thriving on open source technologies. (7)

What's especially remarkable about AI going open source is the speed and scale.

In 2023, an estimated 80% of all AI-generated images, totaling approximately 12.590 billion, were produced using models, services, platforms, and applications built upon Stable Diffusion, an open source diffusion image generation model.(3) In September 2024, the open source platform Hugging Face celebrated 1 million public models hosted.(6)

Traditional software development could take years to match this speed with proprietary alternatives. But AI models can be rapidly improved when thousands of developers worldwide contribute optimizations, training techniques, and innovations. When DeepSeek releases their approach openly, it doesn't just benefit DeepSeek, it becomes a foundation that every other developer can build upon. The same applies to ERNIE, Stable Diffusion, Llama, and the many others entering the market. All free, all open.

This creates a compounding effect that proprietary systems may not be able to match.

While a closed system improves at the rate of its internal team, open source systems improve at the rate of their entire global community. It's not just about having more people working on the problem, it's about having more diverse perspectives, use cases, and innovations flowing back into the core system.

The discovery of solutions to community problems goes right back into the code of the software.

Open source has the freedom and energy to make societal change.

It will be important for countries, cities, and organizations of all shapes and sizes to take ownership over their own AI systems. This concept is often referred to as "digital sovereignty."

Open source fundamentally democratizes who gets to participate in cutting-edge technology development. When AI capabilities are limited to massive computing budgets, only a handful of technology giants get to shape our future. But when those same capabilities become open and accessible, universities, independent researchers, and local startups can contribute breakthroughs.

Instead of competing to build the best private systems, developers can compete to contribute the most valuable improvements to shared systems. While this may seem idealistic, the decision to open software is often more effective because of the ethos. When you're building something that thousands of other developers will use and scrutinize, you naturally build it more thoughtfully.

Open source software tends to be built in modular pieces that connect and disconnect cleanly. The blockchain guys call this "composability." (8) The 3d graphics GLTF standard is highly "moddable" because you don't need to carry unnecessary parts of the file everywhere you go ( looking at you, FBX format ). This mindset is being replicated in projects like the engine O3DE, initially an Amazon project, which introduced the concept of "Gems", modular pieces that allow highly customizable environments. With community plugins readily available, there's usually someone else working on the exact problem you're currently tackling. (9) It creates instant connection and naturally provokes you to contribute back to the project or community.

The AI revolution makes this particularly important because AI systems increasingly impact everyone's daily life. Right now, we're essentially outsourcing crucial societal decisions to private interests. Open source AI creates the possibility of more transparent, accountable, and democratically governed systems.

I also understand that there are legitimate counterarguments to consider. Over the course of many debates, I have been told the following.

Open-source AI can be exploited for misinformation, cyberattacks, and unethical applications without proper regulations. Additionally, while open-source AI thrives on external collaboration, geopolitical tensions could hinder international participation in China's AI projects. (10)

I do agree that these are compelling arguments. I honestly don't think the Chinese released open source to spy on us, but I get the sentiment of the argument.

Ultimately, as an educator and adventurer, the more open the better. I would rather the whole world have access to AI, and our collective efforts should be focused on how to make it work for society.

Open source actually isn't about tech. It's about economics and social systems. It's not about satisfying the customer, it's about creating the components of problem solving tools that the community needs.

My hope is for the emergence of "virtual municipalities," shared technological infrastructure that everyone contributes to and benefits from, similar to how we collectively fund and maintain roads, parks, and public utilities.

In the physical world, we don't expect private companies to own and operate every road, bridge, and water system. We have garbage collection, local elections, water and electricity, and public works. We recognize that certain infrastructure is most effectively managed as a community. The same logic increasingly applies to digital infrastructure. The virtual world needs systems for the well being of its users (both human and automated.)

This argument isn't entirely new, and many are already working on it. Blockchain technologies are experimenting with tokenized networks that reward participants for contributing resources. Open source AI projects are demonstrating how collaborative development can outpace corporate R&D, potentially unlocking new funding through grants and subsidies. If we regulate stablecoins (which seems likely) we'll enable our real economy to integrate with this virtual infrastructure. This could either trigger economic disaster or unlock entirely new forms of virtual economic value.

Ideally, this shift would distribute value among everyone who contributes to the digital space rather than concentrating wealth among a few tech gatekeepers. Instead of competing companies duplicating efforts, we'd see coordinated development toward shared goals. Instead of technologies designed primarily for profit extraction, we'd get tools designed to make humans more awesome.

All that, by putting the code on github and slappin' an MIT license on it.

I still maintain subscriptions to commercial graphics services because paying for established software is often easier for many workflows. But I'm keenly aware that as open source options become more accessible, opportunities multiply for others to capture value or find lower-cost alternatives. If software costs race to the bottom—and everyone becomes a developer creating software on the fly—how will value be created and captured?

Open source isn't perfect.

Many open source tools still lack the polished user experience of their commercial counterparts. Building sustainable funding models for open source development is hard and really hasn't been figured out yet. Coordinating large-scale collaborative projects requires new forms of governance that we're only just starting to figure out.

But what's critical to understand is that, in open source, you're not just consumers of technology built by others. You're a potential contributors to the digital infrastructure that will shape the next several decades. This infrastructure won't be static, it will move and adapt and change. Learning to participate in open source projects, understanding how collaborative development works, and thinking about technology, will be essential skills in this emerging landscape. This is especially true for AI. Think of it as a shared resource like electricity, rather than a private commodity like Coca Cola.

The question isn't whether the open source revolution will happen. It's already here. The question is whether you'll jump in and try it. ( I left some ideas to start your open source journey below. )

Thanks for reading. If you vibe to the ideas I express here, please consider subscribing and sharing. We'll see you next time.

Nye Warburton is technologist and educator from Savannah, GA. This essay was painfully improvised and written with Otter.ai and Obsidian, and augmented with editing and refinement from Llama4 and Claude Sonnet from Anthropic.

For more information visit: https://nyewarburton.com

1. "Between 2022 and 2023, the number of globally available open-source projects saw an average growth of 29%." Source: Sonatype. Open Source Supply, Demand, and Security. https://www.sonatype.com/state-of-the-software-supply-chain/2023/open-source-supply-and-demand

2. "Chinese AI companies are strategically embracing open-source AI to cultivate independent AI capabilities and reduce their reliance on Western technologies." Source: Naik, G. (2025, April 3). China's Open-Source AI Strategy: A Game-Changer in Global AI. https://medium.com/@gitikanaik12345r/chinas-open-source-ai-strategy-a-game-changer-in-global-ai-development-b6fd8864c151

3. "In 2023, an estimated 80% of all AI-generated images, totaling approximately 12.590 billion, were produced using models, services, platforms, and applications built upon Stable Diffusion, an open source diffusion image generation model." Source: Everypixel. (2023, August 15). AI Image Statistics for 2024: How Much Content Was Created by AI https://journal.everypixel.com/ai-image-statistics

4. "Nvidia just dropped Isaac Sim, their mega robot training virtual sim. (*) Again, open source." Source: NVIDIA Isaac Sim - GitHub. https://github.com/isaac-sim

5. "Ninety-nine percent of the world's servers run Linux, and GitHub hosts an eternal playground of frameworks, libraries, tools, and projects with millions of contributors."

Linux Mint | https://linuxmint.com/ - A really approachable and easy to use Linux distro. "Linux for the every day user. " Taking the plunge to go full time with linux was hard, but sooo worth it.

Jan.ai | https://jan.ai/ - A really easy way to jump into local and open AI models. Using Ollama on the back end, you can download and use a variety of models running locally or from API endpoints. It's a great little program.

Hugging Face | https://huggingface.co/ - Essentially the "github of open source machine learning models." Great community, lots of fun things to explore. Just jump in.

(Microsoft Phi, Mistral, Qwen, Llama4 < - all open and available on Hugging Face!)

Deepseek R1 Distill Llama Instruct | https://huggingface.co/jdqqjr/DeepSeek-R1-Distill-Llama-3.2-1B-Instruct - The Deepseek process optimized Llama's several billion instruct model. It runs on a variety of lower GPU set ups, is smaller in size for local downloading, but still excellent for inferencing most text and code. It's my "go-to" open model until the next one comes along!

Gemma (Google's Open Source Model Suite) | https://deepmind.google/models/gemma/ - A collection of lightweight, state-of-the-art open models built from the same technology that powers Gemini models (try it in Googles AI Studio)

Obsidian / Github (and an IDE of your choice) | https://obsidian.md/ and https://github.com/ - I write text in markdown in Obsidian and generally use VS Code if I am working on software. Getting familiar with Github is also recommended. Using AI is about managing prompts. Whether you are writing, or coding, you will need to manage a lot of files.

Anaconda and Python | https://www.anaconda.com/ - If you start to get serious about running things locally ( and you have a computer or VM that has a GPU to play with. ) You should get familiar with virtual environments. I use anaconda, but I am sure there are a bunch of other virtual enviornment set ups to go on this.

• Reframing Crisis as Opportunity, April 13, 2025

• Understanding GPU Tokens, May 4, 2025

• The Future of Work: Embracing Passion, Decentralization, and Automation, July 2, 2024