Nvidia is hiring for a number of roles for a Language Processing Unit team, following its $20 billion Groq deal.

The company has yet to confirm what it plans to do with the technology licensed from Groq, which makes its own LPU AI inference chips, but is expected to announce more at the Nvidia GTC 2026 event next week.

LPUs use on-die SRAM to provide tens of terabytes per second of internal bandwidth, making the chips well-suited for agentic AI inference workloads.

On Christmas Eve, Nvidia spent $20bn to license Groq's tech and hired its leadership team, including CEO Jonathan Ross, as well as other staff. At the time, it was not clear if Nvidia planned to integrate technology from Groq into existing products or develop new product lines.

Multiple previously unreported Nvidia job listings confirm that the company has "an LPU team," which "focuses on creating purpose-built hardware for large language models and generative AI workloads, pushing next-generation AI inference acceleration."

On LinkedIn, a former Groq staffer now at Nvidia notes that they are "working on the low-level communications and systems software libraries of Nvidia's new class of LPU technology."

One job listing, for a 'CPO integration engineer - LPU packaging,' discusses developing silicon photonics (SiPh) and co-packaged optics (CPO) platforms for the LPU. The role involves owning "the definition, design, and qualification of optical packaging architectures for SiPh-based modules, including optical module integration, and thermal/mechanical compatibility."

Groq has yet to use co-packaged optics for its LPUs, but Nvidia has integrated CPO into its networking switches. This role is likely to ensure the LPU works with Nvidia's broader CPO switch strategy, rather than being built into the chip itself.

Whether the first Nvidia-Groq LPUs will be functionally different from Groq's current LPUs is not clear. Groq still exists, just with much of its team gone and its technology licensed to the market's biggest player.

Chinese securities firm GF Securities claims that Nvidia will unveil an 'LPX rack,' featuring a 256-LPU configuration. Others have suggested Nvidia will offer LPUs as on-die units within Feynman GPUs via hybrid bonding.

Former Groq CEO and current Nvidia chief software engineer Ross is set to talk at GTC on the topic of "GPU ♥ LPU: Everything You Wanted to Know," with the conversation taking place after the Monday keynote, where new announcements are usually made.

"With language processing units (LPUs) and GPUs in the same data center, developers can optimize for both ultra-low latency and high throughput," the blurb states. "LPUs excel at token-by-token and interactive workloads, while GPUs excel at massive parallelism and larger batches. This fireside chat helps developers understand the complementary strengths of LPUs and GPUs."

In February, Nvidia CEO Jensen Huang claimed that the company will unveil “a chip that will surprise people” at GTC, with others predicting a silicon photonics announcement.

DCD will be at GTC all next week, alongside sister publication SDxCentral.