The Hidden Abundance of Open-Source AI

ChatGPT, Gemini, or Claude. If you had to pick one large language model (LLM), what would it be? Maybe you lean more towards open-source models like Llama, DeepSeek, Mistral, or Qwen?

In the age of artificial intelligence, a handful of dominant AI players have captured public mindshare and institutional adoption. A trend that is eerily reminiscent of the dot-com era when a small number of tech companies cornered the market and shaped consumer behavior.

Right now, we are in the thick of the AI boom and the technology is maturing at escape velocity. Open-source foundation models are beginning to emerge as dominant players, narrowing the performance gap between themselves and proprietary platforms like ChatGPT. And companies worldwide are beginning to adopt these models.

A recent McKinsey survey shows that among 700 technology leaders and developers across 41 countries, 72% reported incorporating open-source AI models into their tech stacks. Additionally, 76% expect their organizations to increase use of open-source AI over the next several years.

Open-source software has long been a signal of technological progress. It also makes up 99% of the world’s software, according to Martin Woodward, VP of Developer Relations at GitHub. Its development is rooted in the belief that research and innovation should be collaborative, transparent, and accessible. And if you ask Professor Frank Nagle and his colleagues at Harvard Business School, they’d argue that without open-source software and global code-creation networks, companies would spend around $8.8 trillion (3.5 times more) to build the platforms and tools that power their operations.

These open-source tools are also made publicly available with fewer restrictions than their centralized counterparts. Developers can modify, adapt, and redistribute them to meet the specific needs of their organizations. Many teams and agencies cite open-source tools for their performance, flexibility, and significantly lower costs of implementation and maintenance compared to commercial offerings.

Given the growing demand and a thriving ecosystem of open-source development, why do people still default to the same small number of models?

The problem comes down to funding, discovery and distribution.

It is not that alternative models don’t exist. They do, and at scale. But the infrastructure for discovering and adopting open-source AI is broken. Although there are nearly two million open-source generative AI models in circulation today, the AI ecosystem appears far more limited than it actually is.

Open-source creators, even those with technically superior models, are struggling to compete with the likes of Microsoft and Google that have deep pockets and strong customer acquisition pipelines.

The land of forgotten AI models

Platforms like Hugging Face store nearly 1.8 million open-source models that anyone can experiment with, host or finetune upon. But most of these models remain buried in obscurity, not because they lack value, but because they don’t have billion-dollar infrastructure clusters and marketing engines behind them. This is coupled with the fact that users do not have an effective platform to find, evaluate, or use them. For example, OpenRouter analytics show that LLM users continue to default to the same four companies (OpenAI, Google, DeepSeek, and Anthropic) which are driving the majority of traffic.

This lack of visibility has consequences. If no one hosts or uses these models, there is no feedback loop. Without feedback and adoption, contributors lose the incentive to continue improving and maintaining them. The entire ecosystem risks stagnation.

Having a handful of open-source models rise to the top isn’t the same as relying on a handful of closed ones. Any type of open-source model, no matter how dominant, still allows developers to fine-tune, adapt, and build on them, which ultimately fuels a broader ecosystem of innovation. But visibility remains a challenge. Without better infrastructure for discovery and deployment, most developers default to what’s familiar.

For AI to reach its fullest potential, we need to prioritize open and transparent development. It has to be powered in a distributed way, where models can be hosted, accessed, and scaled across independent infrastructure.

This vision is bold, but achievable. Hosting and monetization offers one possible path forward. By decentralizing access to compute and incentivizing model usage and contribution, the open-source ecosystem can create the conditions for long-term sustainability and innovation.

De facto choices leaves the future of AI at risk

The reality is that many developers and organizations default to closed models simply because it is easier. The tooling, documentation, and support around proprietary systems have been carefully optimized to remove friction. Open-source tools need that same level of polish in visibility and performance, and getting there isn’t difficult. What’s standing in the way are artificial barriers imposed by dominant platforms.

What’s at stake is more than just market share. It is the future of AI itself: who controls it, who benefits from it, and how it evolves. Concentrating power in the hands of a few companies limits experimentation, slows progress, and risks embedding systemic bias into the core of the digital economy.

Open-source AI offers an alternative path. But it won’t thrive without community support, visibility, and a renewed focus on discovery and incentives. AI doesn’t have to be a winner-takes-all race, but we need to build the systems and incentives that ensure it isn’t.

This is what Function Network aims to solve. We have built infrastructure that allows model creators to host and monetize their work, while at the same time, giving builders one-click access to a multitude of open-source models. The goal is to make it just as easily accessible, discoverable, and rewarding as our closed-source counterparts.

Function’s testnet is now live on Base. By aligning incentives with model creators, compute providers, and developers, Function is helping ensure that AI remains open for everyone.

Erick Ho is CEO and co-founder of Function Network, a new decentralized AI protocol launching on Base that rewards people to contribute. Prior to this, Erick served as a Senior Software Engineer at Coinbase and advised seed to Series A startups as a Solutions Architect at Amazon Web Services. He specializes in scalable infrastructure and decentralized systems.