Revolutionizing Enterprise AI Security: Nvidia Rubin’s Rack-Scale Encryption Breakthrough

Nvidia recently unveiled the Vera Rubin NVL72 at CES 2026, a revolutionary rack-scale platform that encrypts every bus across 72 GPUs, 36 CPUs, and the entire NVLink fabric. This innovative technology marks a significant advancement in confidential computing, offering security leaders a new way to ensure the protection of complex hybrid cloud configurations.

Instead of relying solely on contractual trust with cloud providers, organizations can now cryptographically verify the security of their infrastructure. This shift is crucial in the face of increasing cyber threats from nation-state adversaries capable of launching targeted cyberattacks at machine speed.

The escalating costs of training frontier AI models, as highlighted by Epoch AI research, present a pressing challenge for organizations. With training costs growing at a rate of 2.4x annually since 2016, billion-dollar training runs may soon become a reality. Despite these escalating costs, the infrastructure safeguarding these investments remains insecure in many deployments.

IBM’s 2025 Cost of Data Breach Report revealed that a significant percentage of organizations have experienced breaches of AI models or applications, with the majority lacking proper AI access controls. Shadow AI incidents, on average, cost organizations $4.63 million, underscoring the urgent need for enhanced security measures.

The disclosure of the GTG-1002 cyberattack by Anthropic in November 2025 shed light on the evolving landscape of cyber threats. This state-sponsored group manipulated Claude Code to execute a large-scale cyberattack with minimal human intervention, showcasing the capabilities of autonomous intrusion agents.

A comparison between the performance of Blackwell GB300 NVL72 and Rubin NVL72 reveals the impressive capabilities of the latter in terms of inference compute, NVLink bandwidth, and HBM bandwidth per GPU. This highlights the industry momentum towards confidential computing, with organizations increasingly adopting this approach to enhance data security.

AMD’s Helios rack offers an alternative approach to Nvidia’s confidential computing technology, prioritizing open standards through the Ultra Accelerator Link and Ultra Ethernet consortia. The competition between Nvidia and AMD provides security leaders with options to tailor their security strategies to their specific infrastructures and threat models.

Security leaders are now leveraging hardware-level confidentiality to bolster zero-trust principles, enabling cryptographic verification of trust. By integrating these technologies, organizations can extend zero-trust enforcement across thousands of nodes without compromising security.

In conclusion, the evolution of cyber threats underscores the critical importance of implementing robust security measures in AI deployments. The deployment of attested infrastructure, such as Nvidia’s Vera Rubin NVL72, is essential for protecting high-value AI investments. Organizations must prioritize security governance, realistic threat exercises, and rack-scale encryption to safeguard their assets effectively. The question facing CISOs is not whether attested infrastructure is worth it, but whether organizations can afford to operate without it in the age of sophisticated cyber threats.