Google Cloud Next ’25: New AI chips, Agent Ecosystem Challenge Microsoft and Amazon

Google cloud It runs an aggressive play to unify its location in the increasingly increasingly artificial intelligence scene, and announces a comprehensive set of new technologies that focus on it.Thinking models“The ecosystems of the agent, and the specialized infrastructure specifically designed to spread AI on a large scale.

In the annual The next cloud conference In Las Vegas today, Google revealed the seventh generation Tensioner Unit (TPU) The name IronwoodThe company claims to offer more than 42 Exaflops from computing energy per pod – which is the most amazing 24 times of superolo computers in the world, El Capitan.

“The opportunity with artificial intelligence is great as it happens.” “Along with our customers, we are running a new golden age of innovation.”

The conference comes in a pivotal moment for Google, which witnessed a great impetus in its cloud work. In January, the company stated that its cloud revenue, Q4 2024 It reached 12 billion dollars30 % increase year on an annual basis. Google CEOs tell active users in Artificial Intelligence Studio and Gemini API It increased by 80 % only last month.

How the new TPUS of Google converts the computing of artificial intelligence with energy efficiency

Google puts itself as the only main cloud provider that contains a “completely improved platform” that is designed from A to Z for what he calls the “era of reasoning”-where the focus moves from typical training to the use of artificial intelligence systems actually to solve problems in the real world.

Google infrastructure star IronwoodAnd, which represents a fundamental shift in the philosophy of chips. Unlike the previous generations that are designed with the balanced Ironwood, Ironwood is specially designed to run the complex AI models after their training.

“It is no longer about the data placed in the form, but what the model can do with data after its training,” Vahdat explained.

Each ironwood contains more than 9000 chips and provides two better energy efficiency than the previous generation. This focus on efficiency is one of the most urgent concerns about artificial intelligence: its huge consumption of energy.

In addition to the new chips, Google opens the huge global network infrastructure for institutional customers through Cloud Wan (Swind Network). This service makes the fiber network worth 2 million miles-which is the same network of consumer services such as YouTube and Gmail-available to companies.

According to Google, the Cloud Wan improves the network performance by up to 40 % while reducing the total cost of ownership at the same percentage in the same percentage compared to the networks run by customers. This represents an extraordinary step for excessive performance, mainly turning its internal infrastructure into a product.

Inside Gemini 2.5: How Google thinks “thinking models” improves AI Enterprise applications

On the side of the program, Google expands the Gemini model family with Gemini 2.5 flashThe cost -effective version of the pioneering artificial intelligence system, which includes what the company describes as “thinking capabilities”.

Unlike the large traditional language models that directly generate responses, these “thinking models” break complex problems through multi -step thinking and even self -thinking. Gemini 2.5 Pro,, Which were launched two weeks ago, are placed in highly used use cases such as drug detection and financial modeling, while the newly announced flash variable adjusts the depth of his thinking based on the immediate complexity of the balance between performance and cost.

Google is also greatly expanding the capabilities of obstetrics with updates to Imagen (To generate photos), Po (video), chirp (Voice), and entry LeriaText model to music. During a demonstration during the press conference, Nenshad Bardoliwalla, Director of Products Management for Vertex AI, explained how these tools can work together to create a promotional video of a music concert, complete with dedicated music and advanced editing capabilities such as removing unwanted elements from videos.

“Only Vertex AI collects all these models, along with third -party models on one platform,” said Bardoliwalla.

Beyond single artificial intelligence systems: How the Google multi -agent ecological system aims to improve the functions of the Foundation’s work

Perhaps the most ascetic ads have focused on creating what Google A calls “Multi -agent eclipse– An environment in which multiple AI systems can operate together through different platforms and sellers.

Google offers ADK Development Group (ADK) This allows developers to create multi -agent systems with less than 100 lines of code. The company also proposes a new open protocol called Agent2agent (A2A), which would allow the artificial intelligence agent of different sellers to communicate with each other.

“The year 2025 will be transitional years, as artificial intelligence moves from answering one questions to solving complex problems through mummified systems,” Fahd predicted.

More than 50 partners have signed to support this protocol, including major institutions program providers Salesforceand servicenowAnd BaitThis indicates a potential manufacture of operating artificial intelligence systems.

For non -technical users, Google improves it Space agent A platform with features such as Al -Wakil Gallery (Providing one offer for available factors) and Agent designer (An interface without a symbol to create dedicated agents). During the demonstration, Google made it clear how the bank account manager can use these tools to analyze customer portfolios, cash flow expectations, and automatically formulate communications to customers – all without writing any code.

From the summary of the documents to driving orders: How the AI ​​agents from Google affects industries

Google also integrates artificial intelligence through it Work area Production suite, with new features such as “Help me analyzeIn the papers, which automatically define visions of data without explicit formulas or pivotal tables, and Voice Overview In documents, which create human -like audio versions of the documents.

The company highlighted five categories of specialized agents, as it sees great dependence: customer service, creative work, data analysis, coding, and safety.

In the world of customer service, Google referred to Driving system of artificial intelligence WendyWhich is now dealing with 60,000 requests per day, and Home Depot.Magic“The agent who provides guidance to improve homes. For creative teams, such as companies WPP Amnesty International uses Google to depict marketing campaigns widely.

The Cloud Ai Contest: How to challenge the comprehensive approach to Google Microsoft and Amazon

Google ads come in intensive competition in the Cloud AI space. I have a depth Microsoft Openai technology integrated through its Azure platformWhile Amazon builds its own Anthropist offers and specialized chips.

“The company’s commitment to providing global infrastructure, models, platforms, and agents; offering an open and multi -fasteious platform that provides flexibility and choice; and based on inter -employment.”

This multi -side approach appears to be designed to distinguish between Google and competitors who may have strengths in specific areas but not the full stack of chips to applications.

Future of the Foundation AI

What makes Google ads particularly important is the comprehensive nature of artificial intelligence strategy, which extends over allocated silicon, global networks, models development, agent framework frameworks, and application integration.

The focus on improving reasoning reflects instead of just the mature AI market training capabilities. While training in excessive models, controlling the main headlines, the ability to spread these models on a large scale has become the most urgent challenge for institutions.

Google’s concentration may indicate an interconnection – allowing systems from different sellers to work together – also to shift from the walled garden styles that feature earlier stages of cloud computing. By proposing open protocols such as Agent2agentGoogle puts itself as the connective tissue in a heterogeneous environmental system of artificial intelligence rather than claiming adopting completely or nothing.

For the technical decision makers of the institutions, these ads provide opportunities and challenges. Efficiency gains promised by specialized infrastructure such as Ironwood TPUS and Cloud Wan can significantly reduce the costs of spreading artificial intelligence on a large scale. However, the rapidly developed mobility of models, agents and tools will require precise strategic planning.

With the continued development of the most advanced artificial intelligence systems, the ability to organize many specialized intelligence agents who work in a concert may become the main difference of AI applications for institutions. When building both components and communications between them, Google bets that the future of artificial intelligence is not only related to the most intelligent machines, but about the machines that can speak effectively with each other.

Daily visions about business use cases with VB daily

If you want to persuade your boss at work, you have covered VB Daily. We give you the internal journalistic precedence over what companies do with obstetric artificial intelligence, from organizational transformations to practical publishing operations, so that you can share visions of the maximum return on investment.

Read with us privacy policy

Thanks for subscribing. Check more VB news bulletins here.

An error occurred.