Google will make Gemini and Agentspace available in on-premises environments from the third quarter. This will enable running Gemini models and applications locally without having to share data with the cloud.
Gemini is the star of the show during Google Cloud Next in Las Vegas. Google announces new additions to the AI arsenal, including Gemini Flash 2.5, a more cost-efficient version of Gemini 2.5 Pro that was launched before the conference.
All Gemini models have in common that they were exclusively available via Google’s cloud servers, but that’s about to change. Google will soon offer the possibility to run Gemini and applications based on it in on-premises environments.
Gemini on-premises
This doesn’t mean Google is completely letting go of Gemini. Google enables the on-premises version of Gemini through its Distributed Cloud platform. This is a Google-managed cloud solution for on-premises and edge, available in connected and disconnected (air-gapped) configurations.
Nvidia has more than a strong influence. Gemini runs on Blackwell systems, such as the DGX B200 and HGX B200, which can be ordered through Google. Thanks to this collaboration, organizations can deploy generative AI without data leaving their own infrastructure. According to Google, the solution offers support for multimodal processing of text, image, audio, and video, as well as support for contexts of millions of tokens and more than a hundred languages.
Through the Gemini API, companies can deploy generative AI without worrying about infrastructure management or model maintenance. Moreover, they can integrate their own data via Retrieval Augmented Generation (RAG), for example to summarize documents or perform sentiment analysis on customer feedback.
read also
What is RAG or Retrieval Augmented Generation?
Agentspace on-premises
Along with Gemini, Agentspace will also be available in the distributed cloud from the third quarter. Companies will be able to use Agentspace on data they prefer to keep on-premises and ask questions via the chat interface. Agentspace supports multimodal queries across data in various formats, such as text, image, PDF, or video. The tool takes access rights into account, which is crucial for compliance in sensitive sectors.
Google links Agentspace to Vertex AI so companies can also build their own homemade AI agents. Thanks to ready-made connectors, you can hook up Agentspace and your AI agents to other systems you use, such as SharePoint and ServiceNow.
Local or in the Cloud
With this announcement, Google Cloud follows a shift in the market towards more locally running AI solutions. Companies show willingness to work with AI but are not always eager to send internal data to the cloud. While cloud may offer advantages in terms of scalability and speed, it is not the preferred option in terms of privacy.
PC manufacturers such as HP and Lenovo are responding to this with homemade AI assistants that run on your PC so you no longer need a cloud subscription, and Microsoft is also looking for ways to make Copilot work locally. Although Google would prefer nothing more than for you to use Gemini from its cloud servers, it finds itself compelled to follow market trends.