End of AI rental: why companies are exchanging expensive APIs for open source, Hugging Face CEO said

Companies are failing to rent artificial intelligence. According to Clem Delange, CEO of Hugging Face, migration to open source AI models is no longer a niche trend — it is a survival strategy. “They don’t want to be trapped in ecosystems that control access, prices and data,” Delange told TechCrunch. The platform, which is now used by half of Fortune 500 companies, has become the epicenter of this movement, hosting thousands of open models that promise autonomy and predictable costs.

Why are companies trading proprietary AI APIs for open source?

The central argument is the lock-in. Supplier APIs like OpenAI, Google, and Anthropic require monthly subscriptions, token limits, and unilateral updates that can quadruple overnight costs. Companies that built products on these APIs realized that they do not control their own infrastructure. “You don’t want your business to depend on a single API key that can be turned off at any time,” Delange warns. The open source alternative eliminates this risk: the model can be downloaded, inspected, adjusted and executed locally or on its own servers.

How did Hugging Face become an open source reference for AI?

The platform functions as a GitHub for machine learning models. With over 500,000 public models and a community of 15 million developers, Hugging Face centralizes the sharing, versioning and testing of architectures such as Llama, Mistral and Stable Diffusion. The company also offers fine-tuning and deployment tools, reducing the technical barrier. Half of Fortune 500 giants already use the platform, not just for consumption, but to contribute to their own open models — a clear sign that the balance of power is changing.

What does open source adoption of AI mean for Brazilian devs?

For technology teams in Brazil, the movement is doubly advantageous. First, it eliminates dollar costs: running locally in Portuguese models avoids spending on exchange conversion and latency of foreign servers. Second, it allows deep personalization for the local market — adaptation to accents, slang and data protection laws such as LGPD. Tools like Ollama, which recently raised US$65 million to facilitate the local execution of models, complement this ecosystem. The Brazilian developer can now implement tailored AI without relying on international gateways.

Impact of open source advancement on the AI market

Migration to open source pushes large suppliers to rethink their business models. Companies like Meta (with Llama) and Mistral French startup already release open versions that compete directly with paid solutions. Hugging Face, by consolidating as the central hub, gains influence over standards, governance and innovation. “We are seeing a decentralization of the power of the AI”, says Delange. If before artificial intelligence was a rented resource, now it is becoming a collectively built heritage.

Questions AI APIs vs. Open Source Common

What is lock-in in AI APIs?

Lock-in occurs when a company becomes dependent on a single API provider, unable to migrate to another solution without high cost or technical rework. In the case of AI, this means losing control over prices, updates and service availability.

Is Hugging Face free?

Yes, the platform is free for basic use and sharing of public models, but offers paid plans for teams that need advanced functionality such as private versioning and dedicated support.

Can small companies also benefit from open source?

Yeah. Smaller and more efficient models, such as versions 7B and 13B of Llama, can run on modest hardware, allowing individual startups and developers to integrate AI at no high cost.