Why Are Companies Abandoning AI Rentals and Turning to Open Source?

A New Wave in the World of AI

The artificial intelligence landscape has undergone a remarkable shift in recent years. Companies are no longer compelled to lock themselves into major model providers and pay steep monthly bills for access to AI capabilities. Instead, a growing number are turning to open-source models that offer greater independence and lower costs. This is what Clément Delangue, CEO of Hugging Face, consistently observes in the day-to-day behavior of the dozens of companies that rely on his platform.

Hugging Face: Infrastructure for the AI Age

Hugging Face has evolved into something akin to GitHub for the AI world — a central repository that allows model developers, research institutions, and companies to freely share and download models and datasets. Its user base has reached record levels, with roughly half of Fortune 500 companies now drawing on the platform's open resources. This widespread adoption is no coincidence; it is the natural outcome of rapid economic and technological shifts that are redrawing the map of the AI market.

The Recurring Pattern: From Subscription to Autonomy

Delangue observes a pattern that repeats itself with companies almost consistently, and it can be summarized in three stages:

• Exploration phase: Companies begin by relying on API access to frontier models such as GPT or Claude — the easiest and fastest options for experimentation.
• Scaling phase: As usage grows, costs accumulate rapidly, and IT teams find themselves facing invoices that strain their budgets.
• Transition phase: The company moves to open-source models run on its own infrastructure, achieving tangible savings while maintaining an acceptable — or in many cases superior — level of performance.

This sequence reflects a straightforward economic reality: renting is convenient at first, but ownership pays off in the long run.

Concerns Over Concentration and Oligopoly

Delangue's concerns extend beyond the commercial dimension. He expresses genuine worry about a future in which a small number of technology companies control the global AI infrastructure. While closed models are often marketed as the safer and more capable choice, he argues that this framing serves the interests of specific companies more than it serves users or society at large. He cites the Anthropic case and the discontinuation of Fable as an example that illustrates the risks of total dependence on the decisions of centralized corporate actors.

Open vs. Closed: The Debate Is Far From Settled

The debate over the future of AI remains open between two camps: one that believes closed models are the most secure and controllable, and another that sees transparency and openness as the true guarantees of innovation and accountability. Yet what Hugging Face's data reveals is that the market has begun quietly voting in favor of the open option — not out of ideology, but out of economic pragmatism and a desire for technological sovereignty.

Ultimately, this may not be a technical issue at its core, but rather one of control, power, and freedom. The companies that recognize this early are the ones that will hold the reins of their own decisions in a future being shaped right now.

✦ بقلم فريق دروب أيديا