DOJ reviews NVIDIA's ~$13B Hugging Face acquisition

The DOJ has opened a review of NVIDIA's planned acquisition of Hugging Face, valued at about $12.9-13B. The central question is whether owning the dominant public hub for model weights, datasets and Spaces would let the dominant AI chip vendor favour its own hardware and software stacks. The deal is expected to close in H1 2027.

Hugging Face is the default distribution channel for open weights. Mistral's 1-trillion-parameter Large 4 checkpoint landed there this week, as do Llama, Qwen and DeepSeek releases. Regulators will likely ask about defaults, inference endpoints, model-card optimisation hints and whether non-NVIDIA accelerators get equal treatment. NVIDIA already ships NIM microservices and Nemotron models through the hub.

The timing adds to scrutiny of NVIDIA as AI's financier. This week it put $3B into Meta's share sale on top of roughly $36B in cloud backstops and up to $105B in OpenAI data-center guarantees. Critics argue that circular financing obscures true demand. A distribution chokepoint added to that financial web is the kind of vertical integration antitrust reviewers notice.

The hub also has its own security problems. Pluto Security found 31 malicious repos across seven accounts hijacking GPUs for cryptomining, with the flagship model downloaded more than 28,000 times before removal. A review is not a challenge. Watch for a second request, any conditions on neutral hosting, and whether the EU or UK open parallel inquiries.