Groq

A $20B Nvidia-Groq licensing agreement turned an AI-inference startup into both a source of Nvidia products and a subject of DOJ scrutiny.

Who they are

Groq is an AI-inference chip and cloud company led by Jonathan Ross, appearing in coverage as an alternative supplier of hardware for running models rather than training them. Its technology and GroqCloud became central to stories about Nvidia’s December 2025 non-exclusive licensing agreement, under which Groq said it would continue operating independently while several senior executives moved to Nvidia.

The recent arc

Coverage reached its high point in 2025Q4 as reports framed Nvidia’s arrangement as a $20B licensing deal, while some accounts described an asset purchase and a leaked Jensen Huang email denied that Nvidia was acquiring Groq. The December reports focused on the agreement’s unusual mix of technology rights, executive departures including Ross, and reported payouts to investors such as BlackRock and Tiger Global.

In 2026, the story shifted from deal mechanics to commercialization, financing, and oversight. Nvidia announced its Groq 3 LPX inference rack in March, and later reported benchmark results for it; Groq subsequently raised funding led by Disruptive and Infinitum, although August coverage reported a $3.5B valuation versus $6.9B before the Nvidia deal. The latest turn is a New York Times report that the DOJ is investigating whether Nvidia sought to skirt antitrust scrutiny in the 2025 transaction.

The tension

The coverage centers on whether Groq can remain meaningfully independent after licensing its inference technology to the dominant AI-chip supplier and losing senior talent to that supplier. Nvidia gains access to Groq’s low-latency inference approach while facing demand for alternatives from customers such as OpenAI, which Reuters reported had explored AMD, Cerebras, and Groq for inference; the arrangement therefore combines partnership, talent competition, and potential regulatory concern.

Why it matters

If Nvidia can productize Groq technology through its own racks while Groq continues to fund and operate GroqCloud, the deal could become a model for absorbing specialized inference capabilities without a conventional acquisition. That outcome is not assured: Groq’s post-deal fundraising and capacity ambitions point to a continuing business, while the reported DOJ inquiry could determine how closely such licensing-and-talent arrangements are examined going forward.

Related: Nvidia · Jonathan Ross · OpenAI · Nvidia announces the Nvidia Groq 3 LPX, an inference server rack featu · Nvidia agrees to a licensing deal with Groq; CEO Jonathan Ross and oth