It's been quite a stretch for Nvidia. Last Thursday the company agreed to buy Hugging Face. This week, it emerged that the US Justice Department has been quietly looking into an earlier deal, the one that brought Groq's leadership in-house.
What the DOJ is looking at
According to a New York Times report picked up by Reuters on Wednesday, the DOJ is investigating whether Nvidia deliberately structured its Groq arrangement to avoid antitrust review.
Quick recap. Last December, Nvidia signed what it called a non-exclusive licence to Groq's inference chip technology. At the same time, Groq founder Jonathan Ross, senior executive Sunny Madra and other leaders moved over to Nvidia. Groq itself carried on as a separate company. Reports have put the value of the deal at anywhere from $17 billion to $20 billion.
On paper, that's a licence plus some hiring, two perfectly ordinary business decisions. The problem, critics argue, is that the end result looks a lot like an acquisition that never had to be filed with regulators. The DOJ reportedly opened its inquiry shortly after the deal was announced and has since sent Nvidia a formal demand for information.
Don't expect fireworks just yet. The Times' sources suggest a fine is possible if investigators find wrongdoing, but forcing Nvidia to unwind the deal is considered unlikely. Nvidia, for its part, called the Groq story "a prime example of the American system working as designed."
Why it matters beyond Nvidia
This "licence and hire" structure isn't unique to Nvidia. Big tech firms have leaned on similar arrangements to scoop up AI talent without triggering a merger review. If the DOJ decides this counts as a stealth acquisition, a lot of future deals will need a rethink, and startup founders planning their exit may find their options narrower than they thought.
Meanwhile, over at Hugging Face
The timing is awkward, because Nvidia has just announced its second-biggest purchase ever. On September 3 it agreed to acquire Hugging Face for $12.93 billion, a figure Bloomberg says includes up to $1 billion in equity-based retention for staff who join Nvidia.
If you write code that touches machine learning, you've almost certainly pulled a model from Hugging Face. Nvidia says the platform has more than 18 million users sharing over 3 million models and 500,000 datasets, with more than 200,000 companies using it.
Jensen Huang has promised that Hugging Face stays open, that it will keep supporting models from every builder, and that Nvidia hardware won't be required to build on or deploy through it. D.A. Davidson analyst Gil Luria compared the deal to Microsoft buying GitHub in 2018 and framed it as a defensive move: better Nvidia owns it than one of the big AI labs, or Google.
Our take
For now, nothing changes for developers. Your `from_pretrained()` calls will work tomorrow exactly as they did yesterday. But with regulators already asking hard questions about how Nvidia absorbs smaller players, the Hugging Face deal is likely to get a much closer look than Nvidia would like. One to watch.
