At Google Cloud Next 2025, the spotlight was on the advancements in the Gemini AI model family, but Google Cloud Platform also unveiled some exciting new infrastructure capabilities.
One major highlight was the introduction of an on-premises version of Gemini, along with Cloud WAN, a managed networking service, and the next generation of Google’s Tensor Processing Unit (TPU) chips. This comprehensive array of AI services positions Google ahead of its main rivals, AWS and Microsoft Azure, as noted by Chirag Dekate from Gartner. However, AI may not be the top priority for IT leaders right now. Many are grappling with other pressing issues in the market, like tariffs. “This year, cost efficiency and other factors are taking precedence in executives’ considerations,” Dekate explained.
With Google Distributed Cloud, organizations can now access Gemini AI models on an on-premises appliance equipped with Nvidia’s Blackwell GPU. This service provides an entire Google Cloud infrastructure stack, ensuring data security through an air-gapped setup, and is distributed via various partners. While GDC previously supported Vertex AI, the upcoming integration of Gemini brings generative AI capabilities, making it suitable for heavily regulated industries.
Additionally, Google will introduce an on-premises version of Google Agentspace, an AI agent service designed to function like a chatbot for enterprise data. Until now, many businesses lacked access to first-party frontier models from companies like OpenAI and Anthropic. “Enterprises were locked out of this technology,” Dekate pointed out. A public preview of Gemini and Agentspace on GDC is set for the third quarter of this year.
Looking ahead, Google’s AI services in the cloud will soon leverage Ironwood, the latest TPU version. This chip comes in two configurations: a 256-chip model and a massive 9,216-chip setup. Promising double the performance of last year’s Trillium TPU, Ironwood allows for the connection of tens of thousands of TPUs for big AI application development. It’s tailored for Google’s Pathways software stack, aimed at streamlining large language model functionality.
On the networking front, Google has made Cloud WAN available now, offering fully managed wide area network services utilizing the robust infrastructure Google has refined over the years. This service covers various scenarios, such as connecting data centers or migrating edge and branch networks. Cloud WAN boasts an open ecosystem with key industry partners like Cisco and Fortinet, providing dedicated, private connections to Google Cloud. Additionally, it facilitates cross-cloud connectivity with other platforms like AWS and Azure, which also have their WAN services.