Nvidia says Groq 3 LPX inference chip is now in production, with Nebius planned as a launch customer before year-end

Nvidia NVDA said its Groq 3 LPX inference chip is now in production and that Nebius has been lined up as the launch customer before the end of 2026, according to a report published Tuesday by Yahoo Finance.

The announcement is the latest milestone in what Nvidia has described as a major, fast-moving bet on Groq technology. The report frames it as an eight-month sprint that began after Nvidia was reported to be pursuing an approximately $20 billion Groq deal, with the company now moving from planning to hardware availability.

Groq 3 LPX is positioned as an “inference” chip, meaning it is aimed at the stage of an AI workload where models generate answers from prompts and not the earlier “training” phase where models learn from large data sets. Inference is increasingly where enterprises are spending to run models at scale, especially once a model is selected and deployed.

By naming Nebius as the launch customer, Nvidia is also providing a concrete path for early deployment. Launch customers typically serve as first reference sites and can help shorten the timeline between chip availability and real-world system integration, even when volume ramp details are not disclosed publicly.

The report does not spell out performance specifications, manufacturing capacity, or pricing for Groq 3 LPX. It also does not provide any breakdown of what portions of Nvidia’s broader infrastructure stack will be used alongside the chip, beyond the production status and the timing expectation for Nebius.

Still, the timing matters for Nvidia’s broader strategy. Over the past year, investors have increasingly focused on whether semiconductor leaders can maintain momentum not only in accelerators for model training, but also in the systems needed to run inference efficiently across cloud providers and enterprise deployments.

Sector context is also important: AI spending is shifting toward “time-to-answer” and total cost of ownership, where inference efficiency can directly affect how many requests a system can handle per dollar. A chip moving into production suggests Nvidia is working to reduce friction for deployments rather than keeping Groq technology in the research or pilot phase.

What remains unclear is how broadly Groq 3 LPX will be rolled out after Nebius. Nvidia did not, in the reporting cited here, disclose forecasted shipment quantities, qualification timelines for additional customers, or whether future configurations will target particular model types, data center power constraints, or specific inference software stacks.