Google doubles down on agentic AI
What happened
- Google made agentic AI the centrepiece of its enterprise pitch at Cloud Next, positioning agents as workflow automation tools. - The company announced partnerships with Merck and BCG, launched two new AI chips, and promoted a multibillion compute deal with Thinking Machines Lab. - Google is tying models to cloud infrastructure, consulting channels and custom silicon to turn AI into repeatable enterprise revenue rather than one-off demos ( ).
Why it matters
Google used its Cloud Next conference on April 22 to put AI agents at the center of its sales pitch to big companies. (blog.google) (usnews.com) An AI agent is software that can plan and carry out multi-step tasks, not just answer a prompt once. Google said its new Gemini Enterprise Agent Platform is built to help companies “build, scale, govern and optimize” large fleets of those systems. (blog.google 1) (blog.google 2) Google tied that software pitch to usage numbers. Sundar Pichai said Google’s first-party models are now processing more than 16 billion tokens per minute through direct customer application programming interface use, up from 10 billion last quarter, and said just over half of Google’s 2026 machine-learning compute investment is expected to go to the Cloud business. (blog.google) The company also paired agents with custom hardware. Google introduced two eighth-generation Tensor Processing Units, TPU 8t for training models and TPU 8i for low-latency inference, and said both are scheduled for general availability later in 2026. (blog.google 1) (blog.google 2) That matters because enterprise buyers have spent the past year moving from chatbot trials to questions about cost, control and scale. Google and Boston Consulting Group said on April 22 that they were expanding their partnership to move clients beyond the artificial intelligence “pilot phase” and into wider deployment with measurable business results. (bcg.com) (blog.google) Google used Merck to show what that looks like in practice. Merck said it will invest up to $1 billion over several years on Google Cloud infrastructure, engineers and Gemini Enterprise licenses to use AI across research, manufacturing, regulatory work and commercial operations. (merck.com) (usnews.com) The customer list was part of a broader argument that Google wants recurring corporate spending, not one-off demos. Google said nearly 75% of Google Cloud customers now use its AI products, and 330 customers processed more than 1 trillion tokens each over the past 12 months. (blog.google) Google is also trying to lock in demand from frontier model makers, not just traditional companies. TechCrunch reported on April 22 that Mira Murati’s Thinking Machines Lab signed a new multibillion-dollar Google Cloud agreement, valued in the single-digit billions, for infrastructure that includes systems built on Nvidia’s GB300 chips. (techcrunch.com) That deal was described as non-exclusive, and TechCrunch said Thinking Machines may still use multiple cloud providers over time. Even so, it showed Google bundling cloud capacity with storage, Kubernetes and database services as it competes with Amazon, Microsoft and Oracle for the biggest artificial intelligence workloads. (techcrunch.com) By the end of Cloud Next, Google’s message was less about a single model than about the stack around it: agents, chips, cloud contracts and consulting partners sold together. That is the business Google spent April 22 trying to prove it can scale. (blog.google) (reuters.com)
Key numbers
- Google used its Cloud Next conference on April 22 to put AI agents at the center of its sales pitch to big companies.
- (blog.google 1) (blog.google 2) Google tied that software pitch to usage numbers.
- Google introduced two eighth-generation Tensor Processing Units, TPU 8t for training models and TPU 8i for low-latency inference, and said both are scheduled for general availability later in 2026.
- (blog.google 1) (blog.google 2) That matters because enterprise buyers have spent the past year moving from chatbot trials to questions about cost, control and scale.
What happens next
- Google used its Cloud Next conference on April 22 to put AI agents at the center of its sales pitch to big companies.
- (blog.google) (usnews.com) An AI agent is software that can plan and carry out multi-step tasks, not just answer a prompt once.
- Google introduced two eighth-generation Tensor Processing Units, TPU 8t for training models and TPU 8i for low-latency inference, and said both are scheduled for general availability later in 2026.
Quick answers
What happened in Google doubles down on agentic AI?
Google made agentic AI the centrepiece of its enterprise pitch at Cloud Next, positioning agents as workflow automation tools. The company announced partnerships with Merck and BCG, launched two new AI chips, and promoted a multibillion compute deal with Thinking Machines Lab. Google is tying models to cloud infrastructure, consulting channels and custom silicon to turn AI into repeatable enterprise revenue rather than one-off demos ( ).
Why does Google doubles down on agentic AI matter?
Google used its Cloud Next conference on April 22 to put AI agents at the center of its sales pitch to big companies. (blog.google) (usnews.com) An AI agent is software that can plan and carry out multi-step tasks, not just answer a prompt once. Google said its new Gemini Enterprise Agent Platform is built to help companies “build, scale, govern and optimize” large fleets of those systems. (blog.google 1) (blog.google 2) Google tied that software pitch to usage numbers. Sundar Pichai said Google’s first-party models are now processing more than 16 billion tokens per minute through direct customer application programming interface use, up from 10 billion last quarter, and said just over half of Google’s 2026 machine-learning compute investment is expected to go to the Cloud business. (blog.google) The company also paired agents with custom hardware. Google introduced two eighth-generation Tensor Processing Units, TPU 8t for training models and TPU 8i for low-latency inference, and said both are scheduled for general availability later in 2026. (blog.google 1) (blog.google 2) That matters because enterprise buyers have spent the past year moving from chatbot trials to questions about cost, control and scale. Google and Boston Consulting Group said on April 22 that they were expanding their partnership to move clients beyond the artificial intelligence “pilot phase” and into wider deployment with measurable business results. (bcg.com) (blog.google) Google used Merck to show what that looks like in practice. Merck said it will invest up to $1 billion over several years on Google Cloud infrastructure, engineers and Gemini Enterprise licenses to use AI across research, manufacturing, regulatory work and commercial operations. (merck.com) (usnews.com) The customer list was part of a broader argument that Google wants recurring corporate spending, not one-off demos. Google said nearly 75% of Google Cloud customers now use its AI products, and 330 customers processed more than 1 trillion tokens each over the past 12 months. (blog.google) Google is also trying to lock in demand from frontier model makers, not just traditional companies. TechCrunch reported on April 22 that Mira Murati’s Thinking Machines Lab signed a new multibillion-dollar Google Cloud agreement, valued in the single-digit billions, for infrastructure that includes systems built on Nvidia’s GB300 chips. (techcrunch.com) That deal was described as non-exclusive, and TechCrunch said Thinking Machines may still use multiple cloud providers over time. Even so, it showed Google bundling cloud capacity with storage, Kubernetes and database services as it competes with Amazon, Microsoft and Oracle for the biggest artificial intelligence workloads. (techcrunch.com) By the end of Cloud Next, Google’s message was less about a single model than about the stack around it: agents, chips, cloud contracts and consulting partners sold together. That is the business Google spent April 22 trying to prove it can scale. (blog.google) (reuters.com)
