Nvidia Raises AI Chip Revenue Forecast to $1 Trillion, Unveils Vera CPU and Groq Deal
At GTC, Nvidia raised its AI chip addressable revenue to $1 trillion through 2027 with split inference architecture leveraging Vera CPUs and a $17 billion Groq IP deal. Nvidia partnered with STMicroelectronics on 800 VDC-to-12V and 6V power conversion architectures for gigawatt-scale AI datacenters, boosting efficiency and thermal performance.
1. GTC Conference Raises Revenue Estimate
At its annual GTC conference, Nvidia increased its addressable AI chip market forecast to $1 trillion through 2027, up from $500 billion previously projected. CEO Jensen Huang emphasized the growing commercial urgency of AI inference workloads as model deployment scales globally.
2. Split Inference Architecture Unveiled
Nvidia introduced a two-stage inference pipeline pairing its Vera CPUs and Rubin GPUs for prefill processing with Groq chips handling latency-sensitive decode tasks. The $17 billion IP licensing deal with Groq underpins this architecture, optimizing compute efficiency for large-scale AI applications.
3. Collaboration with STMicroelectronics on Power Conversion
Nvidia and STMicroelectronics developed new 800 VDC-to-12V and 6V power conversion modules for AI datacenters, targeting gigawatt-scale deployments. These architectures aim to enhance energy efficiency, reduce copper usage, and improve thermal management in high-density server environments.