Semiconductors

Google Launches Ironwood TPU: 24x Faster AI Chip

Ken Metral

09 Apr 2025 — 2 min read

Credit: Alphabet, Inc.

At its Cloud Next conference 2025, Google introduced Ironwood, the latest and most advanced addition to its custom-built Tensor Processing Unit (TPU) lineup. As Google’s seventh-generation TPU, Ironwood marks a pivotal evolution — purpose-built specifically for inference, the process of running and deploying AI models.

Designed with the demands of large-scale inferential workloads in mind, Ironwood will be available later this year for Google Cloud customers in two configurations: a 256-chip cluster and a massive 9,216-chip cluster.

“Ironwood is our most powerful, capable, and energy-efficient TPU yet,” said Amin Vahdat, VP of Google Cloud, “And it’s purpose-built to power thinking, inferential AI models at scale.”

Performance & Specs

According to Google’s internal benchmarks, each Ironwood chip delivers up to 4,614 TFLOPs of peak computing power. Backed by 192GB of dedicated RAM and bandwidth nearing 7.4 Tbps, Ironwood is engineered to handle the complex, high-throughput requirements of modern AI systems.

A standout feature of Ironwood is its enhanced SparseCore, a specialized subcomponent designed to accelerate data processing in workloads like advanced ranking and recommendation engines — the kind that power personalized suggestions on e-commerce and streaming platforms.

Google also emphasized that Ironwood’s architecture is optimized to minimize on-chip data movement and latency, translating into not only better performance but also significant power efficiency gains.

Part of Google’s AI Hypercomputer Vision

Ironwood isn’t just a standalone chip. It will be integrated into Google’s broader AI Hypercomputer infrastructure — a modular, high-performance computing cluster designed for large-scale model training and inference on Google Cloud.

This move is part of Google’s strategic response to the escalating race in the AI hardware space. While Nvidia continues to dominate with its GPUs, tech giants like Amazon and Microsoft are advancing their own solutions — Amazon with its Trainium, Inferentia, and Graviton chips on AWS, and Microsoft with its Cobalt 100 AI processors on Azure.

The Age of Inference

With the explosive growth of generative AI and real-time intelligent applications, inference — not just training — has become a critical performance frontier. Ironwood signals Google’s intent to lead in this domain by offering a highly specialized, cloud-native AI accelerator built for the future of computing.

“Ironwood represents a unique breakthrough in the age of inference,” Vahdat noted. “With increased computation power, memory capacity, networking advancements, and reliability.”

As AI models become more complex and pervasive, Google’s Ironwood TPU could be a key driver in scaling intelligent applications across industries — from search and recommendation to real-time personalization and autonomous systems.

OpenAI Releases o3 & o4-mini with Visual Reasoning

Just two days after unveiling GPT-4.1, OpenAI has dropped a surprise: two entirely new AI models, o3 and o4-mini — and they might just change the game. Introducing o3: The Pinnacle of AI Reasoning OpenAI describes o3 as its most advanced reasoning model yet. It shines in coding, math, and

Figma Files for IPO After $20B Adobe Deal Falls Through

Cloud-based design platform Figma has taken the first formal step toward becoming a public company, confidentially filing for an initial public offering (IPO) in the United States. The move comes more than a year after its planned $20 billion acquisition by Adobe fell through due to regulatory opposition in the

Patreon Takes on Twitch with Livestreaming

Patreon is stepping directly into Twitch territory with the launch of a native livestreaming feature, allowing creators to host live video directly on the platform for the first time. Starting this week, select creators will get early access to the feature, with a broader rollout planned for summer 2025. The

Lyft Acquires FREENOW for $197M to Enter European Market

Lyft has announced its first foray into the European market with a $197 million cash acquisition of FREENOW, a leading multi-mobility platform headquartered in Germany. The deal, which will see Lyft acquire FREENOW from BMW and Mercedes-Benz Mobility, marks a strategic turning point for the U.S.-based ride-hailing company.