NVIDIA has announced a collaboration with Google Cloud to deliver agentic AI capabilities to enterprises. This partnership allows businesses to use the Google Gemini family of AI models within their own infrastructure by using NVIDIA’s Blackwell HGX and DGX platforms along with NVIDIA Confidential Computing, providing enhanced data protection.
Through the use of the NVIDIA Blackwell platform on Google Distributed Cloud, companies can ensure their on-premises data centres meet regulatory and data sovereignty requirements. Sensitive information, such as patient records, financial transactions, and classified government material, can be locked down securely. NVIDIA Confidential Computing also safeguards sensitive code within the Gemini models, preventing unauthorised access and data breaches.
“By bringing our Gemini models on premises with NVIDIA Blackwell’s breakthrough performance and confidential computing capabilities, we’re enabling enterprises to unlock the full potential of agentic AI,” said Sachin Gupta, vice president and general manager of infrastructure and solutions at Google Cloud. “This collaboration helps ensure customers can innovate securely without compromising on performance or operational ease.”
The combination of confidential computing and the Blackwell platform provides technical assurance for enterprises. It ensures that user prompts sent to the Gemini models’ application programming interface, as well as data used for fine-tuning, stay protected and inaccessible to external parties. Model owners are also shielded from unauthorised tampering, creating a dual-layer of protection that promotes innovation while upholding data privacy.
AI agents driving new enterprise applications
This new development comes at a time when agentic AI is significantly changing the enterprise technology landscape, offering more advanced problem-solving features than traditional models.
Unlike earlier AI models that mostly perceive or generate content based on learned patterns, agentic AI systems have the ability to reason, adapt, and make decisions in changing environments. For instance, within enterprise IT support, a standard AI model might provide a troubleshooting guide when asked. However, an agentic AI system could diagnose the issue, apply a fix, and escalate more complex problems without human intervention.
In the financial sector, a typical AI system may detect potentially fraudulent transactions. An agentic AI model, on the other hand, could investigate suspicious behaviour, proactively block fraudulent transactions before they happen, and even adjust fraud detection rules in real-time.
Overcoming on-premises security challenges
Many organisations are already building cloud-based agentic AI applications that use multimodal reasoning, bringing together text, images, code, and more. However, businesses with strict security or data residency needs have not been able to take full advantage of these capabilities — until now.
Google Cloud will be among the first cloud service providers to offer confidential computing for securing agentic AI workloads across cloud and hybrid environments. Powered by the NVIDIA HGX B200 platform with Blackwell GPUs and supported by NVIDIA Confidential Computing, this solution helps protect AI models and sensitive data while achieving high performance and energy efficiency.
This ensures enterprises do not have to choose between innovation and compliance, making it easier to advance their AI projects without risking data or model security.
Enhancing observability and security for agentic AI
Running agentic AI models at scale in production demands strong observability and security measures to guarantee consistent performance and regulatory compliance.
To support this, Google Cloud has introduced a new GKE Inference Gateway designed to improve AI inference workload deployment. The gateway optimises performance through advanced routing and scalability and works with the NVIDIA Triton Inference Server and NVIDIA NeMo Guardrails. This allows for intelligent load balancing, better cost management, and centralised security and governance for models.
Looking forward, Google Cloud plans to enhance observability features for agentic AI by integrating NVIDIA Dynamo, an open-source library that assists in serving and scaling reasoning AI models across AI factories.
NVIDIA will be showcasing more about this collaboration during a special address at Google Cloud Next, alongside various sessions, demonstrations, and opportunities to speak directly with NVIDIA experts.
