Google has stepped on the accelerator in its artificial intelligence strategy. In its cloud Next event, held in Las Vegas, the company has presented a battery of novelties that place its platform as one of the most complete in the market.
Ironwood is the seventh generation of its specialized processor for AI, the well -known TPU (Tensioner Processing Unit). This chip has been designed from scratch for inference tasks, that is, to execute already trained models. According to Google, a single Ironwood rack can offer 42.5 Performance Ex -Flops, 24 times more than the most powerful supercomputer in the world. In addition, it is up to 30 times more energy efficient than the first generation of TPUS.
Google continues to offer NVIDIA GPUS to its cloud customers, but it makes it clear that its TPUS are not just one more option, its hardware is designed to compete.
Gemini 2.5 Pro: “The AI you think before answering”
As for AI models, Google has given more details about Gemini 2.5 Pro, its star model, and has announced a more affordable and fast version called Gemini 2.5 Flash, designed for more everyday tasks such as customer service or the creation of summaries of, for example, a meeting. The main thing about these models in front of the rest, is not only their precision, but their capacity for structured reasoning and multieopa thought, a kind of “the one who thinks before responding.”
In Vertex AI, its platform for developers, creative models have also been integrated as imagine for images, Chirp for audio, I see for video and Lyria for music. During the presentation, we could see a demo that generated a video from a simple image of Las Vegas, music created by AI was added and unwanted objects were eliminated from the plane as if it were Photoshop, but for video.
AI agents
Another of the big bets are the intelligent agents. Google defines them as systems capable of reasoning, collaborating with other agents, even if they are from other companies and execute complex tasks. To facilitate their creation, they have launched an ADK or agent development kit, a kind of kit to mount and deploy agents with less than 100 lines of code. Agent Space was also presented, a platform that allows any employee to create their own agent simply with a conversation, as if they were about configuring a routine in your voice assistant.
In the demo of the presentation, we saw how an alleged bank manager created an agent who analyzed portfolios, generated an audio report and automatically scheduled meetings with customers, without code, only in writing.
In order for agents to work and can be connected to tools and other agents, Google has presented two new protocols, MCP or Model Context Protocol, which allows models to access data in real time, and A2A agent to agent, which facilitates that agents of different platforms can be understood together. This will allow, for example, that a Salesforce agent can speak with another SAP or that the company’s data is always available to all their AI systems regardless of who developed them or where they are stored.
News in the workspace and cybersecurity
Of course, there are also updates on Google Workspace, the company’s office suite. Gemini’s integration is now available for all business users, without the need to pay more, with new functions such as automatic creation of meetings summaries, audio generation from texts and data analysis with natural language in Google Sheets.
As for security, Google has presented specialized agents that analyze malware or alerts in real time, and will be available in on-premise environments through Google Distributed Cloud, even in completely disconnected configurations.
