"Carrot and Stick" to Break Through the AI Competition! Google Cloud Next Conference Unveils AI Intelligent Body "Full Package" with Self-Developed TPU 8t/8i Launching Simultaneously

According to Zhitong Finance APP, as the competition in artificial intelligence (AI) becomes increasingly fierce, Google (GOOGL.US) is attempting to regain its technological edge through a "software and hardware integration" strategy. At the Google Cloud Next annual conference held this week in Las Vegas, Google's cloud computing division not only launched a brand new AI agent building tool to compete with OpenAI and Anthropic but also released the latest generation of its self-developed tensor processing unit (TPU) series, initiating a new round of challenges to NVIDIA's (NVDA.US) dominance in the AI chip field.

Software Layer Enhancement: AI Agents Seizing the Enterprise Automation Entry Point

At the annual conference in Las Vegas, Google Cloud showcased tools that can be used to create AI agents and track their workflows within enterprises, including a dedicated inbox for virtual agents to publish information and progress reports. Meanwhile, Google also updated its Workspace productivity suite and envisioned a future where AI agents will fundamentally change the daily work of ordinary employees.

The core technologies underpinning the current AI boom were pioneered by Google researchers. However, Google is now in fierce competition with leading AI agent manufacturers, vying for enterprise clients eager to leverage AI technology to enhance productivity. Google's capital expenditure for this year alone is expected to reach $185 billion, and investors are hopeful that the company can develop enough new business to support this massive AI investment.

Thomas Kurian, CEO of Google Cloud, stated in a blog post: "We are not providing a single service that can be pieced together; rather, we are building a complete underlying support system for innovation."

The AI programming field is a key focus for Google and is an area where its management is increasingly concerned about falling behind competitors. Several startup founders have revealed that many engineers in Silicon Valley switch between Anthropic's Claude Code and OpenAI's Codex for comparison, rarely considering Google's related products.

To attract developers, Google announced that its Gemini enterprise-level agent platform will add features such as a memory bank and personal memory profiles to address the shortcomings of early AI tools that could not retain historical interaction records. Another new feature, agent simulation, will help developers conduct more comprehensive testing before the tools go live.

Anthropic has extended its business reach into other industries through its Cowork product, and Google is also striving to capture this market. Google launched the Gemini enterprise application, positioning it as "the AI entry point for every employee," allowing users to create agents without writing any code.

Google also introduced a collaboration platform called Projects, which supports employees working together with colleagues and AI agents. This platform can integrate information from Workspace, Microsoft's (MSFT.US) OneDrive, and enterprise chat software, enabling agents to operate in a complete scenario. Other related products launched by Google help clients ensure the safe use of AI agents in industries with compliance requirements In addition, Google has launched a new cybersecurity agent to help customers protect their system security. Although AI models can quickly identify a large number of vulnerabilities, the risk of malicious exploitation is increasingly raising concerns in the absence of a comprehensive protection mechanism.

Hardware Innovation: TPU 8t/8i Debuts, Energy Efficiency Ratio Soars Over One Fold

At the level of computing power infrastructure, Google has simultaneously launched a new generation of TPU chip product line, aiming to further reduce AI inference costs and improve energy efficiency. At the conference, the company stated that the new TPU series will introduce two versions: TPU 8t, which is dedicated to the development and training of AI software, and TPU 8i, which is designed for the operational phase of completed AI services (i.e., the "inference" phase).

In the current industry landscape dominated by NVIDIA, Google has become one of the most successful manufacturers in the field of self-developed AI chips. In recent months, the demand for TPU chips in the Silicon Valley market has continued to rise, and Google hopes to maintain this development momentum with the new generation of products.

The launch of these new products is part of Google's overall strategy to promote lower deployment costs and reduced energy consumption for AI software, while also aiming to enhance service response speed. The new TPU chips are equipped with larger on-chip storage capacity, enabling quick responses that users expect. However, the increasingly complex multi-layer software architecture continues to drive up the demand for computing power.

Mark Lohmeyer, Vice President of Google Cloud Computing and AI Infrastructure, stated: "The core goal is to achieve the lowest possible response latency at the lowest possible cost per transaction. The total volume of transactions is growing significantly, and only by continuously reducing the cost per transaction can we support the large-scale development of AI technology."

Training AI services and software requires systems to quickly process massive amounts of data, uncover data correlations, and build mathematically expressible model patterns. During the inference phase, when running the trained software and services, the advantages of processors with integrated large-capacity storage become even more pronounced.

This design allows AI to operate without retrieving external storage information, thus achieving more immediate responses, particularly effective in scenarios where computers perform multi-step logical reasoning and autonomous learning optimization.

The TPU 8t chips used for training can be deployed in clusters, forming ultra-large-scale computing power systems of up to 9,600 chips. Google stated that when deploying such large computing clusters, power supply has become a major constraint for data centers, so operators need higher efficiency systems to make full use of limited power resources. The performance per watt of TPU 8t has improved by 124% compared to the previous generation, while TPU 8i has improved by 117%.

The performance enhancement is also attributed to the optimization of Google's self-developed internal network technology, significantly enhancing the efficient communication capability between chips. Google officially announced that AI systems built on this series of chips will be fully open for commercial use later this year.

Google also stated that it will continue to provide related services based on NVIDIA chips for customers in need, as NVIDIA products remain the mainstream solution in the field of AI computing. Lohmeyer revealed that Google plans to be one of the first manufacturers to deploy NVIDIA's new architecture hardware in the second half of this year