AI infrastructure was built around a familiar pattern: train a model, deploy it, then respond to individual requests. But now that pattern is changing.
Agentic and reasoning-driven AI can operate continuously, work across long contexts and carry out multi-step tasks. As a result, compute is no longer the only consideration. Networking, memory, power, cooling and data movement all influence how efficiently AI performs at scale. NVIDIA Rubin has been designed around this new reality.
From individual servers to the AI factory
Rather than optimising individual components in isolation, NVIDIA Rubin uses what NVIDIA calls “extreme co-design”. The Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU and Spectrum-6 Ethernet Switch are designed to work together as a single platform.
At rack scale, Vera Rubin NVL72 combines 72 Rubin GPUs and 36 Vera CPUs. The approach effectively changes the unit of compute from the individual server to the rack – and ultimately the data centre. The aim is not simply increased peak performance, but sustained performance and utilisation across demanding, always-on AI workloads.
Making the economics of AI work at scale
As inference becomes a continuous workload, the cost of producing each token becomes increasingly important. NVIDIA states that Vera Rubin delivers up to 10x lower inference cost per token and can train selected mixture-of-experts models using four times fewer GPUs than Blackwell*.
By integrating compute, networking and other infrastructure at rack scale, Vera Rubin is designed to keep resources highly utilised whilst helping address the power, space, cooling and operational demands of expanding AI environments.
This changes the infrastructure conversation for customers. And AI planning needs to consider not only how much compute is available, but how efficiently that infrastructure can support continuous workloads as demand grows.
A new opportunity for partners
For partners, the move towards AI factories creates an opportunity to have broader conversations with customers about infrastructure design, scalability and AI economics.
Understanding how technologies such as NVIDIA Rubin fit into that picture can help partners prepare for next-generation AI projects as customer requirements develop.
TD SYNNEX can support that journey through assessments, integration guidance, financing and training, helping partners build the knowledge and commercial approach needed to address emerging AI infrastructure opportunities.
Ready for the next phase of AI infrastructure?
Explore the TD SYNNEX NVIDIA Hub to learn more about NVIDIA Vera Rubin and the technologies shaping the next generation of AI.
*Based on manufacturer-projected benchmarks released in the official NVIDIA Rubin Platform Architecture Announcement. Actual performance variations depend on specific model architectures and workload configurations.


