Google Cloud Next 2026 kicked off in Las Vegas on Tuesday and the company did not hold back. Sundar Pichai took the stage to announce a new enterprise agent platform, two specialized eighth-generation TPUs, a $750 million partner fund, and a stat that should make every AI company on Earth pay attention: 75% of all new code written at Google is now AI-generated and approved by engineers. Six months ago that number was 50%.

Let that sink in. Three-quarters of the code at one of the world's largest software companies is being written by machines. And Google is not just bragging about it. It is selling the infrastructure to let everyone else do the same thing.

The Gemini Enterprise Agent Platform

The headline product is the Gemini Enterprise Agent Platform, an evolution of Vertex AI that Google is positioning as mission control for the autonomous enterprise. The pitch is simple: build agents, scale them, govern them, and optimize them, all from one place. Agent Studio gives business users a low-code visual builder. The Agent Development Kit (ADK) gives developers a graph-based framework for organizing networks of sub-agents that tackle complex problems together. A new Agent Memory Bank lets agents dynamically generate and curate long-term memories from conversations. Each agent gets its own Agent Identity with enterprise-grade policy controls.

Google Cloud VP Michael Gerstenhaber framed the shift bluntly: the conversation has moved from 'Can we build an agent?' to 'How do we manage thousands of them?' The platform supports multiday autonomous workflows, agent-to-agent orchestration, and what Google calls an Agent Runtime for provisioning new agents at scale. Partners like ServiceNow, Salesforce, Workday, Atlassian, and Oracle already have enterprise-ready agents surfaced inside Gemini Enterprise.

Peter FitzGibbon, SVP at Insight Enterprises, told CRN the quiet part out loud: 'Agentic development has absolutely gone mainstream. There is no more tire-kicking going on like we had in 2024 and 2025. The customers that are leaning into it are leaning into it hard.'

Two TPUs for the Agentic Era

Google is launching its eighth-generation TPUs with a dual-chip strategy. TPU 8t is optimized for training, scaling up to 9,600 chips with 2 petabytes of shared high-bandwidth memory in a single superpod. It delivers three times the processing power of Ironwood and up to 2x better performance per watt. TPU 8i is optimized for inference, connecting 1,152 TPUs in a single pod with 3x more on-chip SRAM. Google says this architecture can concurrently run millions of agents cost-effectively.

The dual approach is telling. Training and inference have always been different workloads, but most chip makers try to be decent at both. Google is saying the agentic era requires purpose-built silicon for each. That is a bet on a future where the number of AI agents running simultaneously dwarfs even the current token volume, which Google says has already hit 16 billion tokens per minute via direct API use, up 60% quarter over quarter.

The $750 Million Fund: Buying the Ecosystem

Then there is the money. Google committed $750 million in new resources and incentives for consulting firms, systems integrators, and channel partners to build on its AI stack. That comes on top of what Google says is $250 billion in committed total addressable market available for partner offerings. Tata Consultancy Services, HCLTech, Cognizant, Capgemini, and Accenture will each get embedded Google forward-deployed engineers. McKinsey, Boston Consulting Group, Deloitte, and Bain get early access to Gemini and other frontier models to provide training feedback. Google wants every major consulting firm on Earth building agents on its platform.

A Rapid Enterprise Migration offering targets companies moving from Microsoft 365 to Google Workspace, promising five times faster migration. That is a direct shot at Redmond while Microsoft deals with its own Copilot adoption challenges.

The 75% Number

The most consequential data point in the entire keynote was almost casual. Pichai said 75% of new code at Google is now AI-generated and engineer-approved, up from 50% disclosed last fall. Google Meet saw 110 million attendees using the 'take notes for me' feature in the last month, up eightfold year over year. A complex code migration was completed six times faster using agents and engineers together than engineers alone could have done a year ago. The team that built the Gemini MacOS app went from idea to native Swift prototype in days using Antigravity, Google's internal agentic development platform.

But here is the awkward part. Google also disclosed that only about 25% of organizations have successfully moved AI into production at scale. Google itself is operating at a level its own customers cannot yet reach. The question is whether this platform closes that gap or whether Google just announced the operating system for a factory floor that most companies do not yet have the workers to staff.

What It Means

Google Cloud Next 2026 is Google doing what Google does best: building platforms. Not models. Not apps. The connective tissue between your data, your goals, and the AI that gets you there. While OpenAI chases enterprise seats and Anthropic builds safety moats, Google is quietly constructing the plumbing that everyone else has to use.

Three-quarters of Google's own code is AI-generated. The eighth generation of custom silicon is shipping. The agent platform is live. And $750 million is being deployed to make sure the consulting firms that advise every Fortune 500 company are building on Google's stack.

Google I/O is May 19. If Cloud Next was the enterprise play, I/O will be the consumer one. Buckle up.