Google Cloud Next 2025 - Key Takeaways
In this blog
Introduction
Last week at Google Cloud Next, Google unveiled a host of new product announcements at their annual conference in Las Vegas. In fact, they detailed 229 new things across AI, infrastructure, networking, security, and more. In this post, we will cover the highlights from the conference and the most impactful updates for enterprise customers.
Gemini 2.5
The first models built on Gemini 2.5 were released several weeks ago but now Gemini 2.5 Pro is available to everyone. This model contains considerable improvement in both "thinking" and transparency in its answer process. After asking a question of the model, users can easily dive into the model's process, uncovering how it determined its answer and what sources were used. The new models also come with the ability to fine-tune how deep the model goes in its research for answers. This allows users to customize the model's approach to answering queries. For complex questions, you could ensure deeper research but for quick and simple tasks, you could target fast responses and lower costs. In addition, Gemini 2.5 models will receive supervised tuning and context caching (an important feature for cost control) in Vertex AI in the coming weeks.
Agent Development Kit
There was a lot of talk about AI agents this year. One of the most important new products coming out of Google Cloud Next is the Agent Development Kit or ADK for short. It is a flexible and modular framework for developing and deploying AI agents. It comes as a Python package that can be installed which allows developers to create their own custom AI agents. A developer simply defines their agent's logic, the tools it has access to, and how information should be processed. Then the ADK provides all the underlying structure to manage state, orchestration, and interaction with the LLM. The ADK is open-source and designed to work across the LLM ecosystem providing a highly flexible way to quickly develop AI agents.
Agentspace
Google Agentspace is Google's Enterprise tool that allows customers to manage and deploy agents across their company. Released in December of last year, the product has been enhanced with new features that simplify the creation of agents that can interact with enterprise data. For example, the new Agent Designer is a no-code UI that allows end users to create agents using just prompts. This democratizes the ability of the workforce to access the powerful automation abilities of AI.
In addition to agent creation, Google also introduced Agent Gallery, for managing your agents, and Agent Garden and Agent Marketplace, for browsing agents created by others and made available to you by your company. Finally, there's Agent Engine, for deploying agents in a fully managed environment without having to worry about managing infrastructure.
Google also highlighted that everything in Agentspace is built with Enterprise security in mind. For example, a user can only access the data they would normally be able to when asking agents to perform tasks. So, if you connect something like Salesforce or Github to Agentspace, the same security controls will apply to each user as they invoke agents that interact with these applications.
BigQuery with AI built in
We saw many new features announced for BigQuery, and they mostly revolve around building in AI. With the ability to store massive data sets, giving users access to AI features directly in the data storage tool makes a lot of sense. One of the most impactful of these is BigQuery Pipelines, now generally available. Data engineers often spend a lot of time cleaning and preparing data for analysis. This process has typically been performed outside of the database itself with other tools that then pipe the cleaned data into the database. With BigQuery Pipelines, users now have access to powerful AI tooling designed to clean and process data right from within the database tool.
Another great example of building in AI came from the new AI Query Engine, which is coming to BigQuery soon. The AI Query Engine allows for "fuzzy" matching by allowing you to search your data with prompts such as "find and return all the pictures in this database containing trees". This is an incredibly powerful and intuitive way to sift through data sets that would otherwise require considerable development effort to accomplish.
Firebase Studio
A brand new product, Firebase Studio is Google's prompt-based development tool that allows for the creation of entire applications via prompt instructions. The recent popularity of similar tools has shown the vast potential of AI in the development of applications and Firebase Studio aims to be Google's product in this space.
Vertex AI Media Studio
For content creators, Google had many good things to show. They have now combined all their content creation products across video, image, speech, and music under the umbrella of the new Vertex AI Media Studio. In Veo2, Google's video generation tool, they showed off the ability to direct the camera via a simple prompt by typing things like "slowly pan to the right" and Veo2 would construct a video with these instructions. Also added was the ability to perform "inpainting", a technique that edits specific pieces of a video via prompts such as "remove the table from the shot" or "add more red flowers to the scene". Finally, Google showed off Lyria, a new text-to-music tool that allows users to easily create high-quality music from a prompt.
Gemini Mobile App and Live API
Last year, Google showed off Project Astra, an implementation of their Live API which streams video from a mobile device to an LLM able to understand and interact with you in real-time. In Las Vegas this year, I got to experience this product firsthand and was absolutely blown away by how good it is. I was inside the Mandalay Bay casino looking for the House of Blues concert venue. I opened the Gemini mobile app and turned on live video. I then simply asked Gemini "How do I get to the House of Blues from here?" It responded: "It looks like you're in the Mandalay Bay casino, next to the Citizens Kitchen and Bar. Continue straight, past the Sportsbook Grill and then you'll see the House of Blues on your right." The directions were spot on. Gemini was able to locate and orient me from the video feed and provide precise directions for navigating within a casino. It was one of the most impressive AI things I saw at the conference, and it was all available from my phone.
Infrastructure
There was a bevy of new hardware offerings announced, including:
- Ironwood - Google's 7th Gen TPU
- Willow - Google's latest quantum computing chip
- A4X VMs - powered by NVIDIA GB200 NVL72, a system consisting of 72 NVIDIA Blackwell GPUs
- H4D VMs - built on AMD's 5th Gen EPICTM processors
- C4D VMs - built on AMD's 5th Gen EPYC processors
- C4 VMs - built on 6th Gen Intel Granite Rapids CPUs
- Availability of NVIDIA Vera Rubin GPUs
- Cluster Director (formerly Hypercompute Cluster)
- GKE Inference Gateway
- Pathways in GCP (multi-host inferencing)
Each of these innovations showcases the rapid pace at which AI is driving hardware scaling. For AI applications, the cloud continues to be an excellent option for enterprises who want to take advantage of the latest available hardware.
Google Distributed Cloud
Google's edge computing platform known as Google Distributed Cloud (GDC) received a major announcement as well: Gemini and Google Agentspace will now be available on both GDC connected and air-gapped. This allows for organizations with the tightest security controls or with low latency use cases to access Google's flagship AI model to perform AI tasks. Customers I have spoken to are very excited about this development and have already been asking us about implementing it.
Cloud WAN
Google announced Cloud WAN, a service that allows enterprises to use the Google secure fiber backbone for their wide-area network. The service boasts 40% faster speeds than using the public Internet for WAN connectivity.
Cloud Interconnect
The new 400G Cloud Interconnect and Cross-Cloud Interconnect, available later this year, were also announced. The current Cloud Interconnect runs at 100Gbs, so this represents a 4X increase in maximum throughput.
Google Unified Security
Another exciting new product was Google Unified Security. To address the sheer number of security products in existence today, this umbrella product brings together threat intelligence, security operations, and Mandiant features into a single product. Designed to be a single pane of glass to search for threats across your entire attack surface, Google is looking to centralize management for security professionals.
Conclusion
These were just a few of the highlights announced at the Google Cloud Next conference. Google continues to grow its offerings and ecosystem as a premier provider of cloud and AI solutions. We have witnessed the tremendous growth of this conference in parallel with the rise in interest and implementation of enterprise AI. WWT is proud to be part of this ecosystem (and a winner of a Google Partner of the Year award). Google Cloud Next 2025 was another example of the rapid pace of cloud evolution and we're excited to continue helping our customers innovate and transform using the power of the cloud.