Groq

A $20B Nvidia licensing agreement moved Groq from an independent AI-inference challenger into Nvidia’s product and talent orbit while it continued raising capital.

Who they are

Groq is an AI chipmaker focused on inference, with LPUs designed to run AI models at low cost. In the coverage, it first appears as a well-funded alternative to Nvidia and later as the source of inference technology and senior technical talent that Nvidia chose to license and recruit rather than wholly acquire.

The recent arc

Coverage accelerated in late 2025 and remained intense into early 2026 after Nvidia agreed to a non-exclusive licensing deal for Groq’s inference technology. The agreement brought CEO Jonathan Ross and other senior executives to Nvidia, while Groq said it would remain independent; reports also described payouts to key executives and investors including BlackRock and Tiger Global. Confusion over whether Nvidia had acquired Groq was a major part of the story, but a leaked Jensen Huang email denied a reported acquisition even as the licensing arrangement was described as worth $20B.

The story then shifted from deal mechanics to productization and survival as an independent company. Nvidia announced the Groq 3 LPX inference rack and said it would sell new Groq chips to AWS, while Reuters reported it was preparing a Groq-chip version for China. Meanwhile, Groq sought up to $650M from existing investors and subsequently announced a $650M raise to expand capacity; ByteDance’s reported work with InnoStar on an inference chip modeled on Groq LPUs underscored the wider influence of its architecture.

The tension

The central tension is whether Groq can remain a distinct inference-platform company after Nvidia licensed its technology and hired much of its senior leadership. Nvidia is both the dominant rival and the route to broader deployment, while customers such as OpenAI have reportedly explored alternatives including AMD, Cerebras, and Groq amid dissatisfaction with some Nvidia inference chips. Groq’s appeal rests on low-latency, lower-cost inference, but its technology is increasingly being absorbed into the ecosystem it was built to challenge.

Why it matters

If Nvidia’s Groq-based systems reach customers through channels such as AWS while Groq builds the capacity it has outlined, Groq’s LPU approach could matter beyond a standalone startup’s sales: it could help define how inference hardware is packaged and deployed. The unresolved question is whether continued fundraising and independent GroqCloud operations can preserve an autonomous business, or whether the licensing deal and executive departures leave the company chiefly as a technology supplier within Nvidia’s platform.

Related: Nvidia · Jonathan Ross · BlackRock · OpenAI · AWS · Nvidia agrees to a licensing deal with Groq; CEO Jonathan Ross and oth