Nvidia launches $20B Groq AI racks, pledges $13B for Hugging Face and proclaims AGI arrival

SentiSense · Published Sep 6, 2026, 3:34 PM ET · Updated Sep 6, 2026, 9:48 PM ET

Nvidia says the AI inference racks built on Groq's LPU designs go live this year following a $20 billion non-exclusive licensing and talent deal, under which Groq itself remains an independent company. Separately, Nvidia agreed on September 3 to acquire Hugging Face for $12.93 billion, the transaction some outlets round to $13 billion and others to $12.9 billion. CEO Jensen Huang also posted that AGI has arrived, crediting OpenAI's new Astra model, a claim OpenAI has not itself made.

Nvidia's inference push moves from paper to production this year. The company says the racks built on Groq's LPU designs will be online in 2026, following a $20 billion deal. One detail matters for reading the deal correctly: Nvidia bought a non-exclusive license to Groq's designs and hired key personnel, including Groq founder and chief executive Jonathan Ross, but Groq remains an independent company, per CryptoBriefing. This is licensed IP and acquired talent folded into Nvidia's Vera Rubin platform, not an acquired subsidiary.

The larger transaction is software. Nvidia agreed on September 3 to acquire Hugging Face, the open-source model and dataset hub, for $12.93 billion: about $11.9 billion to shareholders plus up to $1 billion in equity retention for employees joining the company, per CNBC and Broadband Breakfast. The deal is expected to close in the first half of 2027 pending regulatory approval, and Nvidia has said Hugging Face will stay open to the broader ecosystem. Coverage describing a "$12.9 billion bet to protect the AI empire" is this same transaction rounded differently, not a second commitment.

Chief executive Jensen Huang added the rhetorical flourish, posting "From ChatGPT to o1 to Astra in 4 years. AGI has arrived. Congratulations @OpenAI team" in reaction to OpenAI's release of GPT-6, codenamed Astra. It is worth noting whose claim this is: Huang's, not OpenAI's, and OpenAI has not itself said AGI has arrived. Huang tied it to Astra having been trained on Nvidia silicon, which is as much a positioning statement about NVDA as a technical one.

Read together, the two deals describe a defensive strategy rather than a triumphal one. Hyperscalers building captive inference silicon are the structural threat to Nvidia's margin, and both moves buy positions adjacent to that risk: the Groq license covers the inference-hardware gap, and Hugging Face puts Nvidia at the distribution layer where developers pick models and, by extension, the hardware underneath them. What to watch is whether the Hugging Face deal draws antitrust attention given how central that hub is to open-model distribution, and whether the Groq-derived racks show up in Nvidia's data center revenue line or stay a platform footnote.

Related Stocks

NVDA