Open-Source Models Quietly Changed the Balance of Power in AI

The power draw of today's largest artificial intelligence models is staggering: they require data centers with thousands of specialized chips, each one drawing kilowatts of electricity. Yet the narrative around AI often glosses over this operational reality in favor of breathless promises about the future. The truth is more prosaic but no less transformative: as capable AI models have become freely available for anyone to download and run, the balance of power in the field has quietly shifted.

The moat that was assumed, not proven

In the early days of AI's rapid growth, it seemed like only a select few labs with deep pockets could train advanced models. These labs had access to rare chips and vast pools of talent, creating an environment where capability was seen as scarce and expensive. However, this scarcity narrative has been quietly upended by the release of open-source models that anyone can use. The gap between cutting-edge AI and freely available models is smaller than many thought possible.

Imagine a water tower in a small town: it's crucial for everyone to have access to clean drinking water, but if only one company controls how much water is dispensed and at what price, the rest of the community suffers. Similarly, when AI capabilities are locked behind paywalls and interfaces, companies must rely on those who control the technology. But as open-source models become more prevalent, this reliance diminishes.

Distribution beats secrecy

Open-source AI models spread like any other useful software: through developers sharing code, tools, and solutions. Thousands of people can tweak these models to improve them in ways their creators might never have imagined. This community-driven approach often outpaces proprietary systems because those closest to a problem are usually the best equipped to solve it.

Think about building a house. If you're working alone with limited resources, progress is slow. But if you collaborate with others who share your goals and can contribute unique skills, the project moves forward more efficiently. Similarly, open-source AI benefits from this collaborative spirit, improving along dimensions that might otherwise be overlooked by a single entity.

What the big labs still hold

Leading research labs still have significant advantages: they can afford to train the largest models, maintain robust infrastructure for serving them at scale, and build brand trust that enterprises value. The absolute frontier of AI performance often emerges first in these closed environments. However, this lead is short-lived as open-source releases quickly catch up.

Consider a race where one runner starts ahead but must keep accelerating to stay ahead. Eventually, the gap narrows, making it harder for the leader to maintain their advantage. This dynamic applies to AI: while leading labs still have an edge, that edge shrinks rapidly due to the rapid spread of open-source models.

A more crowded field

The real shift is in who gets to participate in building advanced AI systems. When capable models are no longer a privilege but a commodity, smaller firms and developers outside major research hubs can contribute meaningfully. The center of gravity has expanded, making it harder for any single actor to dominate the entire landscape.

Imagine a garden where only a few had access to seeds initially. As those seeds spread freely, more people could plant gardens of their own. Similarly, as AI models become widely available, innovation spreads beyond traditional centers, fostering a richer ecosystem of ideas and solutions.

None of this guarantees a perfect outcome. Open-source models raise legitimate concerns about misuse and accountability, and the resources needed to train the largest models remain concentrated in few hands. Yet the quiet reality is that more people now have access to powerful AI tools than ever before. Power doesn't shift overnight; it erodes gradually as assumptions are challenged and new possibilities emerge.