DeepMind CEO Demis Hassabis recently disclosed Google’s ambitious plan to merge its Gemini and Veo AI models. This integration is designed to deepen the models’ understanding of the physical world by leveraging each model’s unique strengths. The Gemini model is renowned for its multimodal capabilities, generating audio, text, and images, whereas the Veo model is trained using YouTube’s vast video data to grasp the real-world physics. This move is part of a broader trend towards developing ‘omni’ AI models that can synthesize multiple forms of media seamlessly.
In related developments, Google’s Veo 2 video AI model has introduced exciting features such as automatic object removal, video expansion, and cinematic style presets. These features are designed to enhance video processing capabilities significantly. Additionally, it now supports video transition interpolation between still images. These innovations are accessible through Google Cloud’s Vertex AI platform. Google has also enhanced its Imagen 3 text-to-image model, the Lyria text-to-music model, and Chirp 3’s synthetic speech features, alongside launching Gemini 2.5 Flash for increased efficiency and performance.
Starting in the third quarter of 2025, Google’s Gemini AI models will be accessible on Google Distributed Cloud (GDC), allowing on-premises deployment. This initiative, in collaboration with NVIDIA, will enable these models to run on NVIDIA Blackwell systems, providing optimized performance, heightened security, and compliance with regulatory standards. This on-premises availability addresses issues concerning data sovereignty and volume, ensuring enterprises can utilize AI while maintaining data control and security. GDC offers services like Vertex AI and Google Agentspace, simplifying the development and deployment of AI-powered applications and agents.
Furthermore, the Gemini 2.5 Flash model is designed for efficiency and high performance, optimizing for low latency and cost. It is ideal for real-time and high-volume applications and will be featured in the Vertex AI platform. NVIDIA’s collaboration with Google Cloud also focuses on integrating agentic AI tailored for enterprises, utilizing their Blackwell and DGX systems with confidential computing to secure sensitive data. These integrations aim to enhance enterprise-level AI capabilities, ensuring robust performance and security compliance in various deployment environments.

Collections

Google plans to integrate its Gemini and Veo AI models to enhance their understanding of the physical world by combining their unique capabilities. The Veo 2 model introduces features like automatic object removal and video expansion. These models, along with enhanced versions of Imagen, Lyria, and Chirp, will be available on Google Distributed Cloud from Q3 2025, optimized for NVIDIA Blackwell systems. This move aims to improve AI performance, security, and data sovereignty for enterprises.

Run Gemini AI On-Premises with Google Distributed Cloud and NVIDIA Blackwell