- Ironwood is the seventh generation chip of Google for computing IA, with 3600 times better performance than its first TPU.
- This chip is part of the AI Hypercomputer architecture and is optimized for complex inference tasks such as those in the model Gemini:.
- Together with Ironwood, Google is revamping its Vertex AI platform with models like Gemini 2.5 Flash, integrations with Google Maps, and tools for AI agents.
- Google Agentspace and new features in Workspace boost business productivity with the Artificial Intelligence.
Google has taken another step in its strategy to consolidate its position in the field of artificial intelligence with the unveiling of Ironwood , its new seventh-generation TPU chip designed to maximize the performance of AI applications in its cloud ecosystem. This announcement was made during the Google Cloud Next '25 event, held in Las Vegas, where numerous other innovations aimed at facilitating the mass adoption of AI solutions by businesses were also showcased.
The technology behind Ironwood not only points to the future of AI, but also strengthens Google's position against competitors like Nvidia , offering a powerful and scalable alternative within its own infrastructure. This is all accompanied by a revamped service offering on the Vertex AI platform, improvements to Workspace, and integrations that aim to bring artificial intelligence even closer to the end user.
Ironwood: Google's next-generation TPU chips
Ironwood is the seventh iteration of TPUs created by Google for AI-related calculations . This system is part of its AI Hypercomputer architecture, a combination of hardware , software, and models optimized to run generative AI at scale. In this context, the advancement of chips like the MediaTek Kompanio Ultra also highlights the importance of innovation in the processor sector.
The new processor has been developed for intensive inference tasks , that is, for running pre-trained models, such as those belonging to the Gemini family. According to data provided by the company, Ironwood achieves a 3600-fold performance improvement over the first publicly available TPU and doubles the energy efficiency of Trillium, the previous generation.
Each pod can contain up to 9.216 Ironwood chips, with a theoretical processing power of 42,5 exaflops , far surpassing the El Capitan supercomputer, which supports 1,7 exaflops. This configuration allows for the execution of highly complex AI models in real time and on massive volumes of data. Furthermore, the impact of chip manufacturing on the artificial intelligence space opens up new possibilities for the future.
In addition, the Ironwood chips integrate 192 GB of HBM memory with a bandwidth of 7.4 TB/s , and have cores specialized in sparse operations, common in recommendation or mass classification systems.
Gemini 2.5 Flash and other generative models
A key part of Google's strategy at this event has been the evolution of its model ecosystem, led by the Gemini family . Within Vertex AI, the company's AI development and deployment platform, two new additions are being launched:
- Gemini 2.5 Pro, aimed at in-depth text analysis, interpretation of medical, legal or scientific documents.
- Gemini 2.5 Flash, a lightweight and fast version, designed for cases where cost efficiency and low latency are a priority, such as in customer service or automatic summary generation.
Gemini 2.5 Flash also stands out for its ability to adjust the depth of reasoning based on the type of query or context , allowing companies to fine-tune the user experience while maintaining control over the resources used. Currently, the need for advanced chips is also being felt by other major players in the sector. For example, Microsoft has strengthened its position in artificial intelligence by purchasing large quantities of Nvidia chips.
Beyond text processing, Vertex AI has incorporated new generative models for different modalities:
- Imagen 3, for realistic image generation and editing.
- Chirp 3, which allows you to create natural voices with just 10 seconds of reference audio.
- I SEE 2, with assisted video editing features such as removing elements or changing camera angles.
- lyria, a development that transforms text into 30-second musical sequences.
According to official spokespeople, Google is currently the only company offering fully integrated multimodal generative models on a single platform . The integration of advanced chips is crucial to maintaining this competitive advantage in cloud processing.
Vertex AI: a hub for multi-agent AI
The Vertex AI ecosystem has grown significantly in the last year, increasing its usage twentyfold, according to Google. Today, thousands of companies are already developing AI applications on this platform , and several new features have been introduced for multi-agent systems.
- Agent Development Kit (ADK), an open-source framework for facilitating the creation and deployment of collaborative AI agents.
- Agent2Agent (A2A), a protocol that enables communication between agents designed by different vendors or technologies.
- Agent Garden, a set of ready-to-use examples and tools, including over 100 pre-built connectors to APIs and business flows.
Furthermore, Vertex AI is expanding its integrations with well-known services like Google Maps , thanks to its "grounding" functionality, which refines the responses of location-based models. The option to automatically optimize the model used based on criteria such as speed, accuracy, and cost has also been included. This type of optimization is becoming critical in an environment where the technology industry is constantly evolving.
Agentspace and new capabilities in Google Workspace
Agentspace is Google's bid to bring artificial intelligence to all employees within an organization . It's a unified environment that combines search functions, enterprise assistants, and task automation.
Among its latest developments are:
- Integration with Chrome Enterprise, which will allow you to search for business resources from the navigation bar.
- Agent Assembler, a no-code interface for creating custom agents.
- Deep Research Agent, to generate executive reports after analyzing large blocks of information.
- Idea Generation Agent, designed to organize creative processes or brainstorming dynamics.
Meanwhile, Google Workspace is introducing new AI-powered features :
- "Help me analyze” in Sheets, transforms spreadsheets into automated analyses without formulas.
- "Audio description" for Google Docs, allows you to create spoken versions to consume content while performing other tasks.
- Automatic workflows with Workspace Flows, to systematize routine tasks such as meeting summaries or email management.
- Gemini in Chats, integrating conversational AI directly into messages with commands singles like @gemini.
Google has also focused on the technical support that underpins its entire AI ecosystem . Its private connectivity network, known as Google WAN, has been deployed in more than 200 countries and offers up to 40% faster performance than public internet. This network not only powers massive services like Gmail and Google Photos, but also trains AI models like Gemini. This type of infrastructure is similar to the efforts Apple and Broadcom are making in developing advanced AI chips.
AI-powered infrastructure, security, and science
In terms of security, Google Unified Security (GUS) centralizes the protection of networks, endpoints, and cloud services, enriching data with threat intelligence collected in real time thanks to Gemini.
Within the scientific field, Google Cloud has incorporated models like AlphaFold 3 and WeatherNext into Vertex AI. The former can predict molecular structure with great accuracy, while the latter improves the precision of weather forecasts, benefiting industries such as agriculture and logistics. These advancements demonstrate how AI can be enhanced with two-nanometer chips that significantly improve computational performance.
The deployment of Ironwood represents the consolidation of a comprehensive system for Google Cloud, where AI, custom chips, and high-speed connectivity converge to deliver scalable solutions . These innovations not only strengthen its presence in the enterprise sector but also set a precedent for the future of cloud computing and applied artificial intelligence. Each of the features presented, from natural language processing to audio and image generation, is designed to facilitate the integration of AI into all aspects of corporate life.
Passionate writer about the world of bytes and technology in general. I love sharing my knowledge through writing, and that's what I'll do on this blog, show you all the most interesting things about gadgets, software, hardware, tech trends, and more. My goal is to help you navigate the digital world in a simple and entertaining way.