Nvidia Targets $1 Trillion AI Inference Market with Vera CPU and Groq Partnership

SAN JOSE, California - Nvidia CEO Jensen Huang announced a $1 trillion AI inference revenue opportunity through 2027 at the company’s annual GTC conference.

Huang introduced the Vera CPU and a new inference architecture splitting workloads: Vera Rubin chips handle ‘prefill’-converting user queries into tokens-while Groq-licensed technology powers the ‘decode’ stage that generates responses.

The move signals Nvidia’s strategic pivot beyond training dominance into high-volume, low-latency AI serving-targeting customers like OpenAI, Anthropic, and Meta as they shift from model building to mass user deployment.

Huang confirmed standalone Vera CPU sales are already a multi-billion-dollar business-and previewed the 2028 Feynman architecture, successor to Rubin Ultra.

He also unveiled NemoClaw, an autonomous AI agent platform integrating privacy and safety controls with OpenClaw’s task-execution capabilities.

Shares rose 1.2% following the announcement.