Nvidia targets $1-trillion in AI chip sales as inference demand surges

Nvidia CEO Jensen Huang announced at GTC 2026 that the company sees a $1-trillion revenue opportunity in AI chips through 2027, up from a $500-billion estimate cited in February. The company unveiled a new Vera CPU and an AI system built on Groq technology (licensed for $17 billion in December) to compete in the inference computing market. Huang outlined a two-step inference architecture: Nvidia's Vera Rubin chips handling the 'prefill' stage and Groq chips handling the 'decode' stage. Nvidia is also targeting autonomous AI agents with NemoClaw and previewed the Feynman chip roadmap expected in 2028. The announcements signal Nvidia's strategic pivot from AI training dominance toward large-scale inference deployment as major AI companies shift focus to serving hundreds of millions of users.