Google will let companies run Gemini models in their own data centers

Google’s cloud unit said Wednesday that clients will be able to run its Gemini artificial intelligence models in their own data centers.

Early access to Google Distributed Cloud will be available in the third quarter, Google said in a blog post. The service is targeted at clients that want to use Google’s cloud technology while retaining control of their data.

As part of the announcement, Google said Nvidia will bring Gemini models to the company’s Blackwell graphics processing units, or GPUs. Companies can buy the chips through Google or other channels.

OpenAI and Anthropic, two of Google’s leading rivals in developing large language models, have shied away from providing access in physical data centers because it gives them less control over the quality and speed of the technology. Cohere has an option for customers to deploy models on their own infrastructure, but the AI startup says it’s slower to set up than going through the company or using clouds.

Google’s gesture may be attractive to a new set of potential customers. Many companies, schools and governments still maintain their own data center hardware, although cloud services have become common in recent years. Even customers that adhere to the secret and top secret U.S. government classification levels will be able to use Gemini through an air-gapped version of Google Distributed Cloud that’s disconnected from the internet.

In 2023, cloud infrastructure spending totaled $140 billion, according to technology industry researcher Gartner. Google had 8% of the market in 2023, while Amazon controlled 39% and Microsoft held 23%, Gartner said. Last month, Google announced a $32 billion agreement to acquire cloud security startup Wiz.

“Our commitment to multi-cloud, along with our investments in infrastructure and in AI, are some of the reasons we’re seeing tremendous movement with customers,” Google Cloud chief Thomas Kurian said on a conference call after the Wiz announcement.

Google’s Gemini models can process text, audio and video feeds, with support for more than 100 languages.