Nvidia is extracting more value from its December agreement with Groq by converting the AI chip startup into a buyer of its own systems, a move that underscores the chipmaker's strategy of absorbing competitive threats rather than merely defeating them.
The deal that started it
In December Nvidia paid $20 billion to license Groq's language processing unit technology and hire its founding leadership, including Jonathan Ross and Sunny Madra. The LPU designs have since been folded into Nvidia's product portfolio, while Groq kept its independence and continued running its cloud service, GroqCloud. A $650 million funding round in June gave the startup fresh capital to operate.
The reversal
This week Groq disclosed plans to install Nvidia systems inside its data centers, allowing GroqCloud customers to run workloads on both LPUs and Nvidia hardware. The arrangement turns a company that once positioned itself as a faster, more efficient alternative into a distribution channel for the very rival it sought to challenge.
What analysts see
Futurum Group chief executive Daniel Newman said the surviving Groq entity demonstrates that scaling AI infrastructure is quicker when building atop Nvidia's stack rather than opposing it. Circular Technology's Brad Gastwirth argued the original transaction looks even more strategic in retrospect: Nvidia neutralized a competitive threat while preserving a platform that now funnels compute demand back to its own silicon. "It can absorb the IP, hire the founding talent, invest in the surviving entity, and then sell that entity the compute," Newman said. "Every layer of that sequence deepens the moat."
The broader pattern
D.A. Davidson's Gil Luria noted that the LPU intellectual property was the real prize and that the new partnership fits a wider pattern of Nvidia making calculated moves to entrench its position. "They play chess when everybody else is playing checkers," Luria said.