Opening the Google Cloud NEXT 2026 keynote, CEO Thomas Kurian wasted little time setting the tone. “You have moved beyond a pilot,” he told the packed audience. “The experimentation phase is behind us, and now the real challenge begins: how do you move AI into production across your entire enterprise?”

His answer: a unified, integrated architecture. One in which chips are designed for models, agents are grounded in enterprise data, and infrastructure is secured end-to-end. “You cannot deliver AI by piecing together fragments of disconnected silicon and isolated platforms,” Kurian said. “To drive real value, you need a complete system.”

Sundar Pichai: 6 times more capex, and AI writing three-quarters of Google’s code

Google CEO Sundar Pichai took to the stage to offer a striking window into how the company is eating its own cooking. In 2022, Google invested $31 billion in capital expenditure. This year, that figure is projected to reach between $175 billion and $185 billion, a nearly six-fold increase in four years. Pichai noted that for 2026, just over half of Google’s machine learning compute is expected to go towards the cloud business, a direct signal of how deeply the enterprise opportunity has been internalised.

The internal transformation is already visible. Nearly 75% of all new code written at Google is now AI-generated and reviewed by engineers, up from 50% just last autumn. But Pichai said the shift is moving beyond AI-assisted coding into fully autonomous agent workflows. To illustrate the point, he described a recent code migration project where a system of agents, working across planning, orchestration, and execution roles, completed the task six times faster than would have been possible a year ago.

The efficiency gains are spreading across the business. Google’s marketing teams used AI models to generate thousands of creative asset variations for a recent Gemini-in-Chrome campaign, cutting turnaround time by 70% and delivering a 20% improvement in conversion. On the security side, AI agents now automatically process tens of thousands of unstructured threat reports each month, a task that would have required thousands of analysts, reducing threat mitigation time by over 90%.

Gemini Enterprise Agent Platform: Mission control for the agentic enterprise

The headline product announcement was the Gemini Enterprise Agent Platform, which Pichai described as “mission control for the agentic enterprise.” The platform provides what Google is calling a full-stack system for building, scaling, governing, and optimising AI agents with the same rigour applied to mission-critical enterprise systems.

At its core is Gemini 3.1 Pro, now available in preview and optimised for complex multi-step task orchestration. Alongside it, Google announced Gemini 3.1 Flash for high-fidelity individual assets, a cost-efficient video model for high-volume applications, and a new audio and music model aimed at enterprise and professional use cases. The platform also now supports models from Anthropic, including Claude Opus and Sonnet, a notable acknowledgement of multi-model enterprise reality.

Kurian outlined five pillars underpinning the system: the AI Hypercomputer for infrastructure; the Agentic Data Cloud for enterprise context; Agentic Defence for security; the Agentic Platform for building and governing agents; and pre-built specialist agents ready for immediate deployment.

New capabilities include agent-to-agent orchestration, zero-trust verification with cryptographic agent identities, native Model Context Protocol (MCP) integration, and full Microsoft 365 interoperability, allowing outputs to flow directly into Word, PowerPoint, and other common office formats.

Apple and NASA among the headline partners

Two partnerships underlined the scale of Google’s ambitions. A collaboration with Apple will see Gemini technology power the next generation of Apple Intelligence features, including a more personalised Siri. And in a moment that drew audible reaction from the audience, Google confirmed a partnership with NASA to use Gemini Enterprise agents to power flight readiness systems for Artemis II — the mission that set a new human spaceflight record for distance from Earth.

Among enterprise customers, Norwegian Cruise Line described how Google’s distributed cloud enables AI-powered passenger services to remain operational even when ships are offline. The deployment contributed to a 20% month-on-month increase in record sales and a 60% reduction in production timelines.

The infrastructure imperative

For the digital infrastructure community, the subtext of the keynote is as significant as the announcements themselves. The shift from AI experimentation to production-scale agentic systems is a step-change in compute, data, and connectivity demand.

Kurian’s repeated emphasis on integrated, always-on architecture, agents that respond to real-time events, process data continuously, and operate across distributed environments points to a sustained increase in cloud infrastructure requirements that will ripple across the entire ecosystem.

As Kurian put it in his closing remarks: “The era of the pilot is over. The era of the agent is here.”

Capacity is reporting live from Google Cloud NEXT 2026 in Las Vegas.

RELATED STORIES

New Google Cloud TPUs could be ‘connective tissue’ for agentic AI, CEO says

BICS boosts Google Cloud partnership to deliver global PSTN and cloud communications for enterprises