Google Unveils Ironwood TPU to Power the New Age of AI Inference

Google’s latest innovation in artificial intelligence hardware marks a significant step toward the future of computing, as the company unveils its seventh-generation Tensor Processing Unit, dubbed Ironwood. These custom-built chips are designed specifically for artificial intelligence workloads and signal a major shift toward what Google calls the “age of inference.” Unlike Nvidia’s general-purpose GPUs, Google’s TPUs are application-specific integrated circuits, meaning they are optimized for AI-related tasks rather than a broad range of computations (The Globe and Mail).

Announced to be available to Google Cloud customers in the coming weeks, the Ironwood TPUs represent a leap forward in performance and energy efficiency. According to Google, the Ironwood offers a tenfold improvement in peak performance over its previous TPU v5p model and delivers more than four times better performance per chip for both training and inference tasks compared to the TPU v6e, codenamed Trillium (MSN). These advances make Ironwood Google’s most powerful and efficient custom silicon yet, built to accommodate the surging demand for AI inference, the process by which trained AI models generate outputs or predictions in real-world applications.

While AI model training continues to require immense computational resources, Google believes the industry’s focus is shifting toward inference, as organizations increasingly deploy trained models to perform practical tasks. In this new era, efficiency, scalability, and responsiveness are paramount. AI inference tasks, while less computationally intensive than training, must handle high volumes of real-time data and generate responses quickly to meet the needs of industries adopting AI-driven products and services (The Motley Fool).

Google’s accompanying release of Arm-based Axion virtual machine instances further complements this push. These instances, currently in preview, promise significant performance-per-dollar improvements, helping reduce costs for both AI inference and “agentic AI” workloads. The term agentic AI refers to autonomous systems that can perform sequences of inference tasks, something Google predicts will dominate future AI deployments. This strategic focus positions the company to capture growing market demand for large-scale, cost-efficient AI computing.

A prime example of this trend is the partnership between Google and Anthropic, an AI company working with large language models. Anthropic recently expanded its use of Google’s TPUs for both training and inference. Under this deal, the company will gain access to one million TPUs, enabling it to scale its AI capabilities as it targets $70 billion in revenue and aims to become cash-flow positive by 2028. The superior efficiency of Google’s new TPU architecture was likely central to sealing this collaboration (MSN).

Beyond its hardware advancements, Google’s cloud division stands to benefit substantially from the AI boom. Historically trailing Microsoft Azure and Amazon Web Services (AWS), Google Cloud is now experiencing robust growth, largely driven by AI adoption. In the third quarter, Google Cloud reported revenue of $15.2 billion, a 34% year-over-year increase, and operating income of $3.6 billion, translating to an impressive 24% operating margin. In comparison, AWS’s revenue rose 20% to $33 billion, while Microsoft’s Azure and other cloud services grew 40% (The Globe and Mail).

As organizations move beyond experimentation and begin deploying AI at scale, the demand for inference computing capacity is expected to surge. With more than a decade of AI chip development, Google is well positioned to capitalize on this expansion. The Ironwood TPU, with its blend of speed, efficiency, and scalability, reinforces Google’s ambition to dominate the infrastructure powering the next generation of intelligent applications. By leading the transition into the “age of inference,” Google is not only reshaping the economics of AI workloads but also redefining how cloud computing will evolve in an increasingly AI-driven world (The Motley Fool).

This article was written by the Covertly.AI team. Covertly.AI is a secure, anonymous AI chat that protects your privacy. Connect to advanced AI models without tracking, logging, or exposure of your data. Whether you’re an individual who values privacy or a business seeking enterprise-grade data protection, Covertly.AI helps you stay secure and anonymous when using AI. With Covertly.AI, you get seamless access to all popular large language models - without compromising your identity or data privacy.

Try Covertly.AI today for free at www.covertly.ai, or contact us to learn more about custom privacy and security solutions for your business.  

Works Cited

The Globe and Mail. “Google's Latest AI Chip Puts the Focus on Inference.” The Globe and Mail, 2025, https://www.theglobeandmail.com/investing/markets/stocks/NVDA-Q/pressreleases/36011202/google-s-latest-ai-chip-puts-the-focus-on-inference/.

MSN. “Google's Latest AI Chip Puts the Focus on Inference.” MSN Money, 2025, https://www.msn.com/en-us/money/companies/google-s-latest-ai-chip-puts-the-focus-on-inference/ar-AA1Q622W.

The Motley Fool. “Google's Latest AI Chip Puts the Focus on Inference.” The Motley Fool, 9 Nov. 2025, https://www.fool.com/investing/2025/11/09/googles-latest-ai-chip-puts-the-focus-on-inference/.

Mehta, Ivan. “Google Is Actively Looking to Insert Different Types of Ads in Its Generative AI Search.” TechCrunch, 25 Oct. 2023, techcrunch.com/2023/10/25/google-is-actively-looking-to-insert-different-types-of-ads-in-its-generative-ai-search/.

Vahdat, Amin. “Ironwood: The First Google TPU for the Age of Inference.” Google Cloud Blog, 9 Apr. 2025, https://blog.google/products/google-cloud/ironwood-tpu-age-of-inference/.

“THIS WEEK IN AI: Anthropic Goes to iPhone, AI Programming, Europe’s Rise.” PYMNTS, 3 May 2024, pymnts.com/news/artificial-intelligence/2024/this-week-in-ai-anthropic-goes-iphone-programming-europe/.