Nvidia plots dominance of entire AI data center with combined CPU-GPU strategy

Nvidia is making a strategic shift to become the exclusive chip supplier for AI infrastructure, moving well beyond its historical focus on graphics processors. The company introduced new performance metrics for its Vera Rubin system—a unified GPU and CPU platform—timed to precede AMD's competing announcements at the company's annual conference.

Why CPUs Matter for Modern AI#

The industry's evolution toward more sophisticated AI agents has fundamentally changed the hardware demands of data centers. While graphics processors remain essential for training and executing AI models, central processors have become increasingly critical for managing data flows, networking tasks, and coordinating complex workloads. This shift prompted Nvidia to aggressively market itself as a comprehensive solution provider rather than a single-component vendor.

Vera Rubin's Architecture and Capabilities#

Vera Rubin succeeds Nvidia's Grace Blackwell hybrid superchip and represents a monolithic design—a single integrated circuit rather than multiple chiplets stitched together. Each NVL 72 super chip stacks 36 Vera CPUs alongside 72 Rubin GPUs in a liquid-cooled rack. The architecture claims ten times the token-processing efficiency per watt compared to its predecessor, alongside enhanced memory bandwidth—nearly triple that of Blackwell—a critical advantage as the industry grapples with memory shortages.

Beyond raw performance, Nvidia engineered the system for operational simplicity. The design dramatically cuts the number of cables required to connect components in multi-rack setups, enabling technicians to deploy each rack in minutes rather than hours. Liquid cooling throughout the system further reduces energy consumption for thermal management.

Market Positioning and Competition#

The company has stated that Vera Rubin will enter production in the second half of 2026, with initial deployments at Microsoft, OpenAI, and Oracle. Notably, OpenAI already operates a test rack. Nvidia's moves come as AMD accelerates its own data center ambitions, having grown its CPU market share substantially over the past two years. Both firms are competing fiercely for multi-year contracts with major cloud operators and AI laboratories.

Nvidia's pivot to CPUs relies on ARM-based architecture, chosen for power efficiency, diverging from the x86 chiplet designs that AMD has mastered. Executives argue that monolithic design avoids "a heavy tax on memory bandwidth and data movement" that fragmented approaches impose.