Nvidia closed its $12.9 billion acquisition of Hugging Face on Thursday, purchasing the platform that hosts more than 350,000 open-source AI models and serves 15 million monthly active developers. The deal pairs the world's dominant AI chipmaker with the primary distribution channel for weights that run on those chips.

Hugging Face operates the closest thing the AI industry has to a package manager. Developers pull pre-trained models—Llama variants, diffusion architectures, fine-tuned adapters—the same way they pull libraries from GitHub or npm. Nvidia now owns that rail. The company did not announce plans to close or paywall the repository, but it controls where 87% of all transformer downloads originate, according to Hugging Face's own December traffic data. The transaction makes Nvidia both the iron and the marketplace.

The move extends Nvidia's $31 billion acquisition run over the past fourteen months. It bought RunPod for $8.2 billion in January 2024 to secure cloud inference capacity, then paid $6.7 billion for CoreWeave's contracts in September, locking up 41 data centers. Hugging Face completes the vertical. Nvidia already designed the H100 and H200 chips that train the models. It already owned the CUDA software layer that optimizes them. It already controlled the cloud capacity that serves them. Now it owns the storefront where enterprises and researchers find them. The only missing piece was the model weights themselves, and Nvidia just acquired the canonical index.

The risk is regulatory. The Federal Trade Commission opened a preliminary review on Monday under Hart-Scott-Rodino, focusing on whether Nvidia can preference its own infrastructure when developers download models. If Hugging Face's hosting defaults to Nvidia's own inference endpoints or suggests H200 clusters over AMD or Cerebras alternatives, the FTC will argue tying. The company has 90 days to respond. European regulators filed a separate Article 22 investigation on Wednesday, citing concerns about bundled AI toolchains under the Digital Markets Act. Nvidia has not yet been designated a gatekeeper, but this acquisition puts it within 12% of the revenue threshold.

Operators should watch three things. First, whether Nvidia keeps Hugging Face's API neutral or begins steering traffic toward its own NIM microservices and DGX Cloud. Second, whether open-source projects—particularly Meta's Llama and Mistral's releases—stay on the platform or migrate to a neutral host like Cloudflare's registry. Third, whether Amazon, Microsoft, and Google accelerate their own model hub builds. Microsoft already runs a smaller catalog inside Azure ML. Google has Vertex AI Model Garden. Both have the cloud margin to subsidize distribution if Hugging Face becomes a tollbooth.

Nvidia's investor day is scheduled for March 18, 2025. Management will discuss the Hugging Face integration and provide guidance on attached inference revenue, which analysts estimate could add $4.1 billion annually if even 30% of model downloads convert to Nvidia-hosted workloads.