Join our daily and weekly newsletters to get the latest updates and exclusive content regarding the leading scope of artificial intelligence. Learn more
Google Cloud He aggressively tries to strengthen his position in an increasingly competitive landscape of artificial intelligence. Announced a wide range of up-to-date technologies focused on “Thinking models“Ecosystems agent and specialized infrastructure designed specifically for vast -scale AI implementation.
In his annual Cloud Next Conference In Las Vegas, Google revealed its seventh generation Tensor processing unit (TPU)IN Ironwood. The company claims that it provides over 42 charts of computing power on POD—stunning 24 times stronger than the world -leading supercomputer, Captain.
“The possibility with AI is as great as possible,” said Amin Vahat, Google Vice President and ML Systems and Cloud AI General Director during a press conference before the event. “Together with our clients, we drive a new golden age of innovation.”
The conference takes place at a key moment for Google, which was a significant impetus in its cloud business. In January, the company announced that the cloud revenues in the quarter of 2024. reached $ 12 billion30% enhance in year to a year. Google directors say dynamic users in For learning and API Gemini It increased by 80% over the past month.
How up-to-date Ironwood TPU Google transform AI calculations with energy efficiency
Google is positioned as the only main cloud supplier with a “fully optimized platform” built from scratch for what he calls the “age of inference”, where concentration moves from the model’s training to the actual operate of AI systems to solve the problems with real ones.
The Star of Google Infrastructure is Ironwood, representing a fundamental change in the philosophy of layout design. In contrast to previous generations, which sustainable training and inference, Ironwood was built specifically for conducting sophisticated AI models after training.
“It’s no longer about data entered into the model, but what the model can do with the data after their training,” explained Vahdat.
Each Ironwood capsule contains over 9,000 tokens and provides twice better energy efficiency than the previous generation. Performance concentration applies to one of the most smoking concerns about generative AI: its huge energy consumption.
In addition to up-to-date systems, Google opens its huge global network infrastructure for business clients Cloud Wan (wide -angle network). This service means that Google-Ta fiber optic network, which supplies consumer services, such as YouTube and Gmail-available for companies.
According to Google Cloud, WAN improves network performance by up to 40%, while reducing the total cost of ownership by the same interest compared to the customer’s networks. This is an unusual step for a hyperslase, fundamentally transforming internal infrastructure into a product.
Inside Gemini 2.5: How Google’s “Models of Thinking” improve the AI Enterprise AI applications
On the Google software side, it expands its family of the Gemini model with Flash Gemini 2.5A profitable version of the flagship AI system, which contains what the company describes as “thinking opportunities”.
Unlike customary vast language models that directly generate answers, these “thinking models” distribute sophisticated problems through multi -stage reasoning and even self -reflection. Gemini 2.5 Pro has entered the market two weeks ago and is set in the case of cases with a high sophisticated, such as discovering drugs and financial modeling. At the same time, the newly announced Flash variant adapts the depth of reasoning based on the rapid complexity of efficiency and cost balance.
Google also significantly expands the generative media capabilities with updates Picture (to generate an image), I see (video), Tweet (audio) and introduction of Lyria, a text model for music. During the demonstration during a press conference, NENSHAD Bardoliwialla, VERTEX AI product management director, showed how these tools can cooperate to create a promotional film for a concert, along with non -standard music and sophisticated editing opportunities, such as removing unwanted elements from video clips.
“Only Vertex AI combines all these models, as well as third -party models for one platform,” said Bardolwialla.
In addition to individual AI systems: how the multi -stage Google ecosystem is aimed at increasing the flow of work of the company
Perhaps the most -looking ads in the future focusing on creating what Google calls “Ecosystem of many agents” – an environment in which many AI systems can cooperate on various platforms and suppliers.
Google introduces Agent development set (ADK) This allows programmers to build systems of many agents with less than 100 code lines. The company also offers a up-to-date open protocol called Agent2agent (A2A), enabling AI agents from various suppliers to communicate with each other.
“2025 will be a transition year in which the generative AI moves from answering individual questions to solve complex problems through agency systems,” predicted Vahdat.
Over 50 partnersincluding the main software suppliers for enterprises such as SalesforceIN Service AND Sap, I registered to support this protocolsuggesting a potential industry change towards AI interoperable systems.
Google improves its own for non -technical users Agent’s space Platform with such functions Agent gallery (providing a single view of available agents) and Designer of agents (interface without code to create custom agents). During the demonstration, Google showed how a bank account manager could operate these tools for analyzing customer portfolios, forecasting problems related to cash flow and automatically communication design for customers – all without writing code.
From the summary of documents to the order: how specialized AI Google agents affect industries
Google also deeply integrates AI in the whole Working area performance package, with up-to-date functions such as “Help me to analyze“In sheets, which automatically identify observations from data without clear formulas or rotary tables and audio inspections in documents that create human versions of audio documents.
The company emphasized five categories of specialized agents that relate a significant party: customer service, original work, data analysis, coding and security.
IN Customer service, Google pointed out AI Wendy drive systemwhich currently serves 60,000 orders per day and Home Depot “Magic apron“Agent who It offers guidelines for the improvement of the house. In the case of original teams of companies such as WPP They operate AI Google to conceptualize and create vast -scale marketing campaigns.
Cloud AI competition is intensified: how the comprehensive approach of Google challenges Microsoft and Amazon
Google ads appear among the intensification of competition in AI Cloud space. Microsoft is deep Integrated OpenAI technology on the Azure platformwhile Amazon is building its own Offers powered by anthropic and specialized tokens.
Thomas Kurian, general director of Google Cloud, emphasized “the company’s involvement in providing world -class infrastructure, models, platforms and agents; offering an open multi -lane platform that ensures flexibility and choice; and building for interoperability.”
This multi -produced approach appears to distinguish Google from competitors who may have strengths in specific areas, but not full piles from systems to application.
The future of AI Enterprise: Why Google’s “Models of thinking” and interoperability for business technology
What makes Google ads particularly significant is the comprehensive nature of the AI strategy, including custom silicon, global networks, creating models, agents’ frames and application integration.
The focus on the optimization of inference, and not only the possibilities of training, reflects the maturing market of artificial intelligence. While training has always dominated the headers, the effective implementation of these models on a vast scale becomes a more urgent challenge for enterprises.
Google pressure on interoperability – enabling systems from various suppliers to cooperate – can also signal a departure from brick garden approaches, which characterized earlier phases of cloud processing. By proposing open protocols such as Agent2agentGoogle is positioned as connective tissue in the AI heterogeneous ecosystem, and does not require adoption.
These ads present opportunities and challenges for technical decision makers of enterprises. Profits from performance promised by specialized infrastructure, such as Ironwood TPU and Cloud Wan, can significantly reduce the costs of implementing vast -scale artificial intelligence. However, moving in the rapidly developing landscape of models, agents and tools will require careful strategic planning.
As these more sophisticated AI develops, the ability to organize many specialized AI agents working in a concert can become a key distinction between the AI Enterprise implementation. By building both components and connections between them, Google bet on that the future of artificial intelligence is not only smarter machines, but also machines that can talk effectively with each other.