Google bets on agents

Published April 23, 2026 by The Daily Scout

- Google Cloud presented AI agents as the centrepiece of its enterprise monetisation strategy at Cloud Next. - Nvidia said customers can build “AI factories” using A5X instances, scaling to nearly one million Rubin GPUs. - The message shifts selling from raw models to managed, long‑running workflow platforms, changing where engineering effort concentrates. (reuters.com) (blogs.nvidia.com)

Why it matters

Google used its Cloud Next conference on April 22 to tell customers it wants to sell AI agents, not just AI models. (cloud.google.com) Chief Executive Thomas Kurian said Google’s new Gemini Enterprise Agent Platform is designed to “build, scale, govern, and optimize agents,” folding model tools from Vertex AI into a managed platform with orchestration, security and DevOps features. (cloud.google.com) Google paired that software pitch with new workplace features: Agent Designer, an Inbox for agent activity, long-running agents, Skills and Projects inside the Gemini Enterprise app. (cloud.google.com) An AI agent is software that can keep working across multiple steps instead of answering one prompt and stopping; Google’s pitch is that companies now need tools to monitor, secure and coordinate those longer jobs. (cloud.google.com) That changes where cloud vendors try to make money. Selling access to a model is one business; selling the software layer that connects models to company data, approvals, logs and other systems is another. (cloud.google.com) Google backed the message with infrastructure claims. Kurian said nearly 75% of Google Cloud customers are using its artificial intelligence products, and that its first-party models now process more than 16 billion tokens per minute through direct customer application programming interface use, up from 10 billion last quarter. (cloud.google.com) Nvidia used the same event to frame the hardware side of that strategy. In a post published April 22, it said customers will be able to build “AI factories” on new A5X bare-metal instances powered by Vera Rubin systems. (blogs.nvidia.com) Nvidia said those A5X systems can scale to 80,000 Rubin graphics processing units in a single-site cluster and to 960,000 Rubin GPUs across multiple sites, using ConnectX-9 networking and Google’s new Virgo data-center fabric. (blogs.nvidia.com) (cloud.google.com) Google’s own networking post said Virgo was built because traditional general-purpose networks are “reaching their breaking points” as model sizes and computing demands grow. (cloud.google.com) The result is a more bundled sales pitch: Google is offering models, agent software, security controls, workplace tools, chips, networking and partner integrations as one stack for companies that want AI systems running continuously inside the business. (cloud.google.com 1) (cloud.google.com 2) The test now is whether enterprises buy that full stack in production. Google’s own keynote said the “Agentic Enterprise” is already “in-production,” and Cloud Next was built around proving it can sell the software and infrastructure to run it. (cloud.google.com)

Key numbers

Nvidia said customers can build “AI factories” using A5X instances, scaling to nearly one million Rubin GPUs.
(reuters.com) (blogs.nvidia.com) Google used its Cloud Next conference on April 22 to tell customers it wants to sell AI agents, not just AI models.
In a post published April 22, it said customers will be able to build “AI factories” on new A5X bare-metal instances powered by Vera Rubin systems.
(blogs.nvidia.com) Nvidia said those A5X systems can scale to 80,000 Rubin graphics processing units in a single-site cluster and to 960,000 Rubin GPUs across multiple sites, using ConnectX-9 networking and Google’s new Virgo data-center fabric.

What happens next

Google used its Cloud Next conference on April 22 to tell customers it wants to sell AI agents, not just AI models.
In a post published April 22, it said customers will be able to build “AI factories” on new A5X bare-metal instances powered by Vera Rubin systems.
Google’s own keynote said the “Agentic Enterprise” is already “in-production,” and Cloud Next was built around proving it can sell the software and infrastructure to run it.

Sources

Quick answers

What happened in Google bets on agents?

Google Cloud presented AI agents as the centrepiece of its enterprise monetisation strategy at Cloud Next. Nvidia said customers can build “AI factories” using A5X instances, scaling to nearly one million Rubin GPUs. The message shifts selling from raw models to managed, long‑running workflow platforms, changing where engineering effort concentrates. (reuters.com) (blogs.nvidia.com)

Why does Google bets on agents matter?

Google used its Cloud Next conference on April 22 to tell customers it wants to sell AI agents, not just AI models. (cloud.google.com) Chief Executive Thomas Kurian said Google’s new Gemini Enterprise Agent Platform is designed to “build, scale, govern, and optimize agents,” folding model tools from Vertex AI into a managed platform with orchestration, security and DevOps features. (cloud.google.com) Google paired that software pitch with new workplace features: Agent Designer, an Inbox for agent activity, long-running agents, Skills and Projects inside the Gemini Enterprise app. (cloud.google.com) An AI agent is software that can keep working across multiple steps instead of answering one prompt and stopping; Google’s pitch is that companies now need tools to monitor, secure and coordinate those longer jobs. (cloud.google.com) That changes where cloud vendors try to make money. Selling access to a model is one business; selling the software layer that connects models to company data, approvals, logs and other systems is another. (cloud.google.com) Google backed the message with infrastructure claims. Kurian said nearly 75% of Google Cloud customers are using its artificial intelligence products, and that its first-party models now process more than 16 billion tokens per minute through direct customer application programming interface use, up from 10 billion last quarter. (cloud.google.com) Nvidia used the same event to frame the hardware side of that strategy. In a post published April 22, it said customers will be able to build “AI factories” on new A5X bare-metal instances powered by Vera Rubin systems. (blogs.nvidia.com) Nvidia said those A5X systems can scale to 80,000 Rubin graphics processing units in a single-site cluster and to 960,000 Rubin GPUs across multiple sites, using ConnectX-9 networking and Google’s new Virgo data-center fabric. (blogs.nvidia.com) (cloud.google.com) Google’s own networking post said Virgo was built because traditional general-purpose networks are “reaching their breaking points” as model sizes and computing demands grow. (cloud.google.com) The result is a more bundled sales pitch: Google is offering models, agent software, security controls, workplace tools, chips, networking and partner integrations as one stack for companies that want AI systems running continuously inside the business. (cloud.google.com 1) (cloud.google.com 2) The test now is whether enterprises buy that full stack in production. Google’s own keynote said the “Agentic Enterprise” is already “in-production,” and Cloud Next was built around proving it can sell the software and infrastructure to run it. (cloud.google.com)