Breaking

NVIDIA Highlights Confidential Computing for Enterprises Running AI on Sensitive Data

September 22, 2026 at 15:33 EDT

Security
Infra & Chips
Sales & Business

NVIDIA is sharpening its enterprise AI pitch around Confidential Computing, arguing that companies can bring AI to sensitive data while keeping both business records and proprietary models protected during execution. The emphasis connects NVIDIA's secure GPU roadmap with VAST Data's AI infrastructure work, where agentic systems need fast access to enterprise context without turning private data into a new exposure point.

Sep. 22, 2026 · NVIDIA

Secure AI Where The Data Lives

NVIDIA is positioning Confidential Computing as the trust layer for enterprises that want AI on sensitive data without exposing records, prompts or model weights during execution.

98%

up to unsecured performance in NVIDIA Blackwell tests

Rubin GPUs in a Vera Rubin NVL72 rack-scale system

Vera CPUs in the same secure rack-scale domain

Measured overhead stayed in single digits on tuned Blackwell tests

The visual below turns the reported 1% to 8% impact range into proportional stacked blocks.

smallest tested drop

largest tested drop

The performance story is not zero cost. It is that a carefully configured secure stack can keep the overhead small enough for production inference planning.

How the trust boundary works

Sensitive data

→

Verified TEE

→

Protected inference

→

Attested result

Why enterprises care

Regulated teams get a path to use private records, prompts and model weights in AI systems while limiting exposure to cloud operators and shared infrastructure.

What still needs proof

Trust depends on hardware generation, CPU TEE support, drivers, attestation operations and workload shape. Poor configuration can erase much of the performance story.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.