Five-Minute Focus on Google Next Highlights! AI and Multi-Agent System Innovation Summary

On April 9, 2025, the highly anticipated Google Cloud NEXT 2025 conference officially kicked off a three-day innovation feast for the global tech community. Microfusion Technology invites you to explore this year’s Google Next event, which achieved significant progress in AI and multi-agent systems, once again solidifying Google Cloud’s position as a leading enterprise-grade AI platform. Microfusion Technology has organized seven key areas to bring you up to speed on the major updates in just five minutes!

The Era of Google AI Supercomputers is Here! TPU Gen 7, A3 Mega, and Pathways Are All Online!

Google Cloud emphasized its leadership in AI infrastructure, introducing the AI Hypercomputer architecture, which integrates TPUs, GPUs, high-speed storage, and optimized networking.

• AI Hypercomputer:
  • The seventh-generation TPU Ironwood offers exceptional performance for AI workloads, with 4 times the computing power and 6 times the HBM capacity compared to the sixth-generation TPU Trillium.
  • vLLM for TPUs supports inference on Compute Engine, GKE, Vertex AI, and Dataflow, and supports custom compute classes in GKE for simultaneous use of TPUs and GPUs.
  • Pathways on Google Cloud, an ML runtime for large-scale training and inference workloads, is now available with Jax.
  • Cluster Director for GKE supports the orchestration of large AI VM cluster deployments across regions, with topology-aware scheduling.
  • Support for NVIDIA A3 Mega VM (H100) provides more flexible LLM fine-tuning and deployment capabilities.
  • Dynamic Workload Scheduler supports TPU and A3 Ultra.

Gemini 2.5 Debuts! Faster, Stronger, More Media Creation Capabilities

Google Cloud continues to invest heavily in AI model innovation, launching Gemini 2.5 Pro and Flash, as well as an expanded suite of generative AI media models.

• Gemini 2.5:
  • Introduces the next-generation “thinking models” Gemini 2.5 Pro and Gemini 2.5 Flash, now available on Vertex AI.
  • Gemini 2.5 Pro is a SOTA model for complex reasoning, coding, and mathematics. It performs exceptionally well in multiple benchmarks and ranks first on the LM arena leaderboard.
  • Gemini 2.5 Flash is a fast, efficient thinking model optimized for speed and cost.
  • Live API (Preview) for Gemini: A production-ready API for real-time, interactive audio experiences with enhanced features. Supports long-lasting, resumable sessions, multi-voice audio output, timestamped transcription, dynamic instruction updates within sessions, and robust tool integration.
  • Vertex AI Model Optimizer (Experimental): Leverages Google’s unique understanding of Gemini to automatically route queries to the most performant model and tool based on customer preferences for quality, speed, and cost.
• Generative AI Media Models:
  • Lyria: A new text-to-music feature that allows customers to create 30-second instrumental clips on Vertex AI in seconds. Currently in Private Preview.
  • Veo 2: Text-to-video and image-to-video model with new editing features and camera presets (such as pan-left/right and drone-style shots). Also includes advanced editing features like interpolation, video restoration, and outpainting. This transforms Veo from a generation tool to a comprehensive video creation and editing platform.
  • Chirp 3: Supports 10-second audio generation and understanding model, with new audio generation and understanding capabilities.
  • Imagen 3: The highest quality text-to-image model, with significant improvements in generating finer, richer-lit images with fewer artifacts. Improved image generation and restoration capabilities, particularly providing a more natural and seamless editing experience when removing objects.

Google Agentspace: Your Work-Exclusive AI Assistant is Here!

Google Cloud is committed to promoting the development and application of AI Agents, launching platforms and tools such as Google Agentspace and Vertex AI Agent Builder.

• Google Agentspace: Aims to provide enterprise employees with convenient AI assistants to help them search, interact, and perform actions at work.
  • Seamless integration with Chrome Enterprise: Allows employees to quickly access and search enterprise resources directly from the Chrome search bar.
  • Agent Gallery provides a single interface for employees to discover and use various AI agents: including agents from Google (such as NotebookLM, Idea Generation Agent, and Deep Research Agent), internal teams, and partners.
  • Agent Designer: A no-code interface where employees can design their own AI assistants to automate daily work tasks based on their needs. Supports the open Agent2Agent protocol.
  • Supports Partner Agents: Customers can easily browse and purchase AI agents from partners like Accenture and Deloitte.
  • Deep Research agent: Leverages Gemini’s multimodal intelligence to help employees transform scattered data into rich, actionable insights and reports.
  • Idea Generation agent: Helps employees generate and develop impactful ideas.
  • Supports Grounding in Google Maps: Ensures that agents’ responses are based on accurate and up-to-date location information.
  • Integration with AlloyDB: Enables Google Agentspace to search structured data.
• Agent Development Kit (ADK): Introduced a new open-source Agent Development Kit (ADK) and runtime Agent Engine to help developers build complex multi-agent applications. ADK is launched in Python and will support more languages in the future.
• Introduced a new open Agent2Agent protocol: Aims to unify the fragmented multi-agent ecosystem and enable collaboration between agents built across different frameworks.

Gemini and Firebase Studio Accelerate the Development and Deployment of AI-Driven Applications

Google Cloud provides developers with powerful tools and platforms to simplify the building and deployment of AI applications.

• Introduced Firebase Studio (currently in Preview): A cloud-based, agent-driven development environment that includes everything full-stack developers need to quickly and securely create production-quality, AI-driven applications. Gemini Code Assist agents can be accessed through Firebase Studio.
• Gemini Code Assist introduces agents and provides more coding assistance interfaces.
  • Code Assist agents help with the following tasks: generating new software, migrating code, implementing new features, performing code reviews, and generating unit tests.
  • Gemini Code Assist tools: Brings partner technologies (such as Snyk and Atlassian) into the IDE.
  • Coding assistance in Android Studio via Gemini Code Assist Enterprise, with enterprise-grade security and privacy.
  • Gemini 2.5 now powers Gemini Code Assist for personal projects.
• Gemini Cloud Assist (Public Preview) introduces new features to simplify cloud operations and management through AI.
  • Design and deploy application infrastructure through natural language.
  • Gemini-guided investigation workflows for fast troubleshooting.
  • Cost and utilization optimization with AI assistance.
  • Enhanced Gemini chat with richer context and the ability to perform domain-specific tasks.
  • Deeper integration with many Google Cloud products (including Storage Insights, Cloud Observability, and Firebase).

AI Integration into Looker, Revolutionizing Data Applications

Google Cloud’s innovations in databases and data analytics aim to help enterprises manage and utilize their data more effectively.

• Looker Conversational Collaboration: Expanded capabilities for conversational analytics, allowing developers to seamlessly integrate natural language queries into any application or workflow, including visualizations, formulas, and automatic presentation generation, as well as LookML Code Assistants to simplify the code creation process.
• New Generative AI Capabilities for AlloyDB: Integrated with Google Agentspace, allowing the use of Gemini’s advanced reasoning and Google-quality search to query and search data in AlloyDB.
• Expanded Oracle and Google Cloud Services: Oracle Base Database Service, global Exadata X11M availability, and plans to add 11+ regions in 2025, totaling over 15 regions.
• SQL Server modernization: Database Migration Service supports SQL Server to PostgreSQL migration (for Cloud SQL and AlloyDB), with low-downtime data migration and a rule-based conversion engine enhanced by specially trained Gemini models.

Security AI-Driven Security Package

Google Cloud emphasized its leadership in cloud security, introducing innovative solutions like Google Unified Security and leveraging AI to enhance security operations.

• Introduced Google Unified Security: An integrated security solution that combines unparalleled threat visibility, faster threat detection, continuous virtual red team exercises, the most trusted browser, and Mandiant’s expertise, all powered by Google AI and operating on a global-scale data network.
• Security Agents & Gemini are transforming security operations:
  • Alert triage agent (Preview): Performs dynamic alert investigations and provides judgments.
  • Malware analysis agent (Preview): Autonomously identifies security issues and recommends whether code is malicious.
• Advanced features of Google Security Operations: