Meta just quietly signaled the biggest shift in its AI strategy since Mark Zuckerberg declared open source was the future. The first AI models built under Alexandr Wang, the 27-year-old former Scale AI CEO who now runs Meta's AI division, will launch as proprietary. Open source versions will come "eventually."

That single word tells you everything. "Eventually" in Silicon Valley means "if it still makes strategic sense after we have extracted maximum competitive advantage." Meta wants to keep "some pieces proprietary" and ensure the models "don't add new levels of safety risk," according to Axios. The safety framing is convenient. The business logic is transparent.

Here is why this matters. Meta's Llama models became the default open source alternative to GPT and Claude. Developers built entire ecosystems around them. Startups raised funding on the assumption that Meta would keep releasing competitive models for free. The open source AI community treated Meta like a reliable utility. This is the first crack in that assumption.

The timing is not accidental. Google just dropped Gemma 4 under Apache 2.0, completely open, running locally on iPhones. Google is eating Meta's open source lunch while Meta pivots to proprietary. If you are a developer choosing between a model you can use right now with zero restrictions (Google) and one that might become open source "eventually" (Meta), the decision is not close.

Alexandr Wang's appointment was already controversial. He built Scale AI into a $14 billion company by selling human-labeled training data to the same AI labs he now competes against. The conflicts of interest write themselves. Now his first move is to lock up Meta's models behind a proprietary wall, even temporarily. This is not the Meta AI playbook developers were promised.

The deeper problem is precedent. If Meta can walk back its open source commitment for one model generation, it can do it for every generation. The entire value proposition of Llama was "we give you the model, you build on it, and the ecosystem makes Meta's platform stronger." That flywheel only works if developers trust the supply. One "eventually" chips at that trust in ways that are hard to repair.

Watch Google closely. Every day Meta delays open sourcing Wang's models is another day Google's Gemma ecosystem grows. The open source AI throne is up for grabs, and the company that sat on it just stood up.

First reported by Axios.