Groq, Inc.

Groq designs the Language Processing Unit (LPU), a deterministic, SRAM-based inference-optimized chip architecture, and operates GroqCloud, a hosted inference service for large language models.

Groq built its LPU thesis around avoiding the HBM and CoWoS packaging bottleneck that constrains GPU-based inference, relying instead on large on-chip SRAM and a mature process node. That positioning helped it raise capital rapidly, culminating in a $750M round at a $6.9B valuation in September 2025. The story changed materially on December 24, 2025, when Nvidia announced a non-exclusive licensing agreement for Groq's inference technology, reported at roughly $20 billion, alongside Nvidia hiring Groq's founder and president; Groq said it continues to operate as an independent company under a new CEO. By March 2026, Nvidia had unveiled a Samsung-manufactured Groq 3 LPU rack product slated for Q3 2026 production, to be deployed at neocloud Nebius alongside Nvidia's own silicon. For our packaging thesis, Groq is notable less as a standalone equity story now and more as evidence that Nvidia is moving to control both the GPU/CoWoS-bound compute layer and a non-HBM inference architecture that had marketed itself as an escape from that exact packaging bottleneck. We hold this at low conviction: what we are tracking now is a consolidation event, not an independent growth thesis.

Groq's core technology and key leadership, including founder Jonathan Ross and president Sunny Madra, have moved to Nvidia, raising the question of what remains at independent Groq and what an investor is actually underwriting going forward. The reported roughly $20B deal figure has not been confirmed by either company, and real economics to Groq's cap table are unclear. Reporting indicated lawmakers were scrutinizing the Nvidia-Groq deal as of early 2026, which could affect deal completion or structure. Groq's non-HBM, non-CoWoS architecture avoided one packaging bottleneck but must still prove out at Samsung's newer 4nm node and against Nvidia's own in-house inference roadmap, now that Nvidia effectively controls both.

Confirmation or denial of the reported roughly $20B licensing figure by Nvidia or Groq in an official financial disclosure. Delay or cancellation of the Nebius Groq 3 rack deployment relative to Nvidia's stated 2026 timeline. Any signal that independent Groq is being wound down entirely rather than continuing to operate.

Groq, Inc. appears here as a research record. Coverage does not disclose or establish investment availability, relationships, permission or allocation.