Q&A: Nvidia exec on how ‘confidential computing’ can secure AI agents - Computerworld

As artificial intelligence (AI) advances, particularly with the rise of agentic AI, data security and privacy have become critical concerns. In a recent interview, NVIDIA's senior director of high-per... Read original →

NVIDIA’s integration of confidential computing into Blackwell GPUs shifts AI security from perimeter-based to hardware-rooted trust. This triggers a cascade: CPU vendors like Intel and AMD must deepen GPU-CPU TEE interoperability, while cloud providers overhaul hypervisors for hardware-isolated AI workloads. Regulatory pressure—from the EU AI Act to U.S. healthcare privacy laws—is making in-use encryption non-optional for regulated sectors, raising operational costs for laggards. Google Cloud will likely fortify its Confidential VMs around TPUs, while Apple may leverage 3nm custom silicon to lock down agent execution on-device. Within 18 months, confidential computing will become table stakes in enterprise GPU procurement, redirecting EUV capacity toward secure accelerators and spurring demand for third-party TEE validation services.