Nvidia turns Groq into customer after $20 billion deal

Groq's decision to deploy Nvidia systems in its data centers completes a strategic arc that began with a $20 billion licensing deal.

Groq will deploy Nvidia systems in its data centers, giving GroqCloud customers access to Nvidia GPUs alongside its signature LPUs — converting a former rival into a customer.

"The fastest way to scale in AI infrastructure is to build on Nvidia, not against it," Daniel Newman, CEO of Futurum Group, said.

The partnership marks a sharp reversal from where the companies started. In December, Nvidia struck a $20 billion deal with Groq to license its LPU technology and hire its senior leadership, including founder Jonathan Ross and president Sunny Madra. Nvidia has since incorporated LPUs into its product lineup. Groq remained independent, raised a $650 million funding round in June, and continued operating GroqCloud.

The arrangement deepens Nvidia's moat in AI infrastructure. "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."

Brad Gastwirth, global head of research and market intelligence at Circular Technology, said the deal "potentially neutralized part of a competitive threat while preserving Groq as a growing platform that can now drive additional demand back toward Nvidia." He noted that Nvidia doesn't need all AI workloads to run exclusively on its chips as long as it captures at least some of the business.

Gil Luria, head of technology research at D.A. Davidson, said the LPU technology was the real prize for Nvidia in the original transaction. While many specialized AI cloud providers — known as neoclouds — were already buying Nvidia's AI chips, this new partnership fits a broader pattern of the chipmaker making strategic moves to strengthen its position. "They play chess when everybody else is playing checkers," Luria said.

The strategic logic behind the deal

Nvidia's dominance in AI accelerators has made it difficult for challengers to gain traction. Its data center segment has grown into a multi-billion-dollar-per-quarter business, driven by demand for GPUs used to train and run large language models. Groq's LPUs, designed for fast and efficient inference workloads, were among the most credible alternatives — offering lower latency and higher throughput per watt for certain tasks compared to Nvidia's general-purpose GPUs.

By licensing the technology and hiring its leadership, Nvidia absorbed the intellectual property while keeping the company alive as a distribution channel. The arrangement also gives Nvidia a foothold in the inference market, where Groq's LPUs had carved out a niche for low-latency workloads. GroqCloud customers will now have access to both Nvidia GPUs and Groq's LPUs, potentially making the platform more attractive to enterprises seeking flexibility in their AI infrastructure choices.

What it means for the AI chip landscape

For other AI chip startups — including Cerebras, SambaNova, and others — the deal signals that Nvidia is willing to use its balance sheet to neutralize threats rather than simply out-compete them on performance. The pattern mirrors how Nvidia has historically handled competitive pressure: acquire or absorb the technology, then integrate it into its broader product lineup.

For investors, the implications are twofold. Nvidia gains a new revenue stream from selling systems into Groq's data centers, while eliminating a competitor that had positioned its LPUs as a faster, more efficient alternative. The $20 billion upfront cost is substantial, but the strategic value — removing a credible challenger and converting it into a customer — could pay dividends for years as AI inference workloads continue to scale across enterprise and cloud deployments.

This article is for informational purposes only and does not constitute investment advice.