INVESTMENT NOTE
Groq
Making real-time AI fast and economical.
THE PROBLEM
As AI moves from experimentation into everyday products, inference—the work required to serve every answer—has become a critical bottleneck. Developers need responses that are fast, predictable and economical enough to run at enormous scale.
THE APPROACH
Groq built its Language Processing Unit specifically for that job. Deterministic, software-scheduled execution and on-chip memory reduce the delays and variability common in general-purpose GPU systems, while GroqCloud makes the hardware available through a familiar developer API. It is a vertically integrated path from specialized silicon to production inference.
WHY THIS TEAM
Founder Jonathan Ross designed core elements of Google’s first TPU. After Groq’s December 2025 technology-licensing agreement with NVIDIA, Ross, president Sunny Madra and several team members joined NVIDIA while Groq remained independent. CEO Adam Winter and CFO Matt Eng now lead a team with experience across Meta data centers, xAI, Microsoft and enterprise software—a strong blend of deep systems knowledge and commercial execution.
MOMENTUM
NVIDIA integrated Groq technology into its Vera Rubin platform as the Groq 3 LPX inference accelerator. In June 2026, Groq raised $650 million to expand its inference cloud. The company reports 13 data centers, more than five million developers and trillions of tokens processed each week; those operating figures are company-reported.