Google Cloud Next ’25: New AI chips and agent ecosystem challenge Microsoft and Amazon

Google Cloud It is an aggressive game to strengthen the position in the view of an increasingly competitive artificial intelligence, announces a broom series of new technologies “Thought models“The agent ecosystems and specialized infrastructure designed specifically for large-scale AI placements.

Annually The next conference Today, in Las Vegas, Google revealed the seventh generation Tensor Processing Section (TPU) named OmasyMore than 42 exaphlops per pod-pod claimed by the company are 24 times more powerful than the world’s leading supercomput. Captain.

“It’s great as the opportunity with AI,” he said. “With our customers, we strengthen a new innovation age.”

The conference comes to the main moment for Google, which seeks considerable speed in the cloud. In January, the company said that the company Q4 2024 cloud revenues Has reached $ 12 billionThe year of 30% per year increases. Google leaders say active users Learn and Gemini API only increased by 80% in the last month.

Google’s new Ironwood TPUS varies with AI calculation power effectiveness

Google places itself as the only key cloud provider that “is a fully AI-optimized platform”, which is called “inference age” – to use model training to use model training to solve real world problems.

The star of Google’s infrastructure ads OmasyWho represents a fundamental turn in chip design philosophy. Unlike the previous generations of balanced training and previous generations, it is specially established to launch complex AI models after the iron tree.

“It’s not about the information on the model, but also what the model is taught,” Vahdat said.

Each Ironwood Pod consists of more than 9000 plugs and provides two-fold power effectiveness from the previous generation. This focus is one of the most current concerns related to the efficiency, the generative AI is: its great energy consumption.

In addition to the new chips, Google opens a mass global network infrastructure to its customers Cloud WAN (Extensive Regional Network). This service is available for enterprises, which provides consumer services such as Google’s 2 million miles – YouTube and Gmail.

According to Google, Cloud Wan, compared to customer-controlled networks, increases the network activity by 40% by reducing the same percentage value in the same percentage. This is an unusual step for a hyperscaler, essentially converts the internal infrastructure into a product.

As part of the Gemini 2.5: How Google ‘Thought Models’ Enterprise Improves Application

On the program side, Google, Gemini Model Family expands Twins 2.5 FlashAn effective version of the flagship AI system described as “thinking skills”.

Unlike traditional major language models that respond directly, these “models” these “thinking models” break up in complex problems by reflecting very steps. Gemini 2.5 ProIt is placed for high-complex use as high-complex and financial modeling, which began two weeks ago, while the newly declared flash option regulates the main depth based on the fast complexity to balance the performance and value.

Google significantly expands generative media capabilities with updates Picture (for image formation), I see (video), Chicken (audio) and application Liriafrom text-music model. During the demonstration during the press conference, Product Management Director Nenshad Bardoliwalla showed a promotional video and complex editing capabilities to create unwanted elements from video clips of these vehicles.

“Only Vertex AI leads all these models to a platform to a single platform,” said Bardoliwalla.

Outside of Single AI systems: Google’s multi-agent ecosystem aims to enhance businesses

Perhaps the most forward-looking announcements aimed at creating Google callers “Many agent ecosystem“- an environment where many AI systems can work together between various platforms and sellers.

Google presents a Agent Development Kit (ADK) This allows developers to build multiplied systems with less than 100 code. The company also offers a new open protocol called AI Agent2Agent (A2A), which allows you to connect with each other.

“2025 will be the year to answer the general questions of the Generative AI, to respond to the solution of complex problems with agent systems,” Vahdat was forecasted.

More than 50 partners have signed this protocol, including large enterprise software Righteousness, Serviceenowand Leakoffers a potential industry to change towards interconnected AI systems.

Google enhances it for non-technical users Agent Platform with features like Agent gallery (to ensure a single view of existing agents) and Agent designer (No code interface for creating special agents). During a demonstration, Google showed how to analyze the customer’s portfolios of a bank account manager, to analyze cash flows problems and automatically communicate customers to customers.

From the document summaries to driving orders: How does Google affect the industry specialized AI agents industry

Google also integrates the AI ​​on its own Workspace Productivity kit, with new features “Help me to analyze“On the sheets automatically identify data information from data without open formulas or pivot schedules and Audio views In documents that create audio versions of documents as a human.

The company stressed five categories of adoption: customer service, creative work, data analysis, coding and security.

In the Customer Service World, he pointed to Google Wendy’s AI drive systemNow 60,000 orders daily and holding the home depot “Magic apron“Agent offering the improvement management of the house. For companies for creative groups WPP Google’s use AI to conceptualize and produce marketing campaigns on a scale.

Cloud AI contest is strengthening: How to make Google’s comprehensive approach to Microsoft and Amazon

Google’s ads come in competition in the cloud AI space. Microsoft is deep Openai technology integrated along the Azure platformWhile Amazon is established Anthrop-working victims and specialized chips.

Director General of Google Tomas Kurian, the company’s “world-class infrastructure, models, platforms and agents” to offer a very cloudy platform that provides comfort and choice; and interacting. “

This very bent approach is designed to distinguish opponents in specific areas in specific areas, but from competitors differ from fully accumulated rivals from chips.

The future of the enterprise AI: Why Google’s ‘Think Models’ and interacting for work technology

Google’s ads are especially important, special silicone, global networking, model development, agency development, agent frames and integration integration integration are a comprehensive nature of AI strategy.

It only reflects an adult AI market, not only training opportunities, but focused optimization. Whenever the larger models are designed, the ability to place these models effectively with the scale of these models becomes more preparation for businesses.

Google’s accent to Google, which allows you to work together from different vendors, can move away from walled garden approaches that characterize the previous stages of the cloud calculation. By offering open protocols such as Agent2agentGoogle places itself as a connecting tissue in the heterogeneous AI ecosystem, rather than demanding everything.

The enterprise provides both opportunities and difficulties for technical decisions. Promised Efficients promised by the Infrastructure such as Ironwood TPUS and Cloud WAN can significantly reduce the costs on a scale of AI. However, the rapidly developing scenery, agents and vehicles of the models will require careful strategic planning.

As it continues to develop more complex AI systems, the ability to orchestrate multiple specialized AI agents working in the concert may be the main distinguish between the enterprise AI application. When building both components and connections, Google has not only smarter machines of the AI’s future, but only bets on machines that can be able to speak effectively with each other.