Cloud Next ’26: Google doubles down on AI agents, security, and infrastructure at massive scale

A keynote stage at Google Cloud Next ’26 showcases AI innovation, highlighting Gemini agents, advanced TPUs, and next-generation cloud infrastructure driving enterprise transformation.

SAN FRANCISCO — If last year’s cloud conference was about proving generative AI works, this year’s Google Cloud Next makes one thing clear: Google is now focused on operating AI at industrial scale—and turning that capability into a competitive advantage for enterprises.

At the center of the announcement is Google Cloud’s accelerating momentum. The company revealed that its first-party AI models are now processing over 16 billion tokens per minute, a sharp rise from 10 billion just one quarter ago. That surge isn’t just a vanity metric—it reflects a structural shift in how businesses are embedding AI into daily operations.

More tellingly, Google expects over half of its machine learning compute investment in 2026 to go toward its cloud business, signaling a strategic pivot: AI is no longer experimental infrastructure—it’s core utility.

The rise of the “agentic enterprise”

The most consequential theme at Cloud Next ’26 is the transition into what Google calls the “agentic era,” driven by its Gemini ecosystem.

Enterprises are no longer asking whether they can build AI agents. The new challenge is scale and orchestration.

Google’s answer is the Gemini Enterprise Agent Platform, positioned as a full-stack system for building, managing, and governing thousands of AI agents simultaneously. Think of it less as a tool and more as a mission control layer for enterprise AI.

In practice, this reflects a real shift happening across industries. A large financial services firm, for example, might deploy:

• Compliance agents scanning regulatory updates
• Risk agents modeling exposure scenarios
• Customer service agents handling millions of interactions

Individually, these agents are manageable. At scale—hundreds or thousands—they become operationally chaotic without centralized governance.

Google’s platform aims to solve that complexity by connecting data, workflows, and human oversight into a single system. Early traction suggests strong demand: paid monthly active users for Gemini Enterprise grew 40% quarter-over-quarter in Q1.

The takeaway for businesses is immediate: AI strategy is shifting from model selection to systems architecture.

AI becomes both weapon and shield in cybersecurity

Security has long been a differentiator for Google, but this year’s announcements reflect a more urgent reality: AI is escalating both cyber threats and defenses.

Google unveiled a new AI-powered cybersecurity platform that integrates its threat intelligence capabilities with solutions from Wiz. The highlight is Wiz’s AI Application Protection Platform (AI-APP), designed to secure applications across code, cloud, and runtime environments.

This matters because modern attacks increasingly target AI pipelines themselves—from poisoned training data to prompt injection exploits.

A practical example: consider a retail company running AI-powered pricing engines. A compromised model could manipulate pricing logic at scale, causing financial losses before detection. Traditional security tools aren’t designed for this layer. AI-native protection is.

Inside Google, the impact is already measurable. Its internal Security Operations Center uses AI agents to triage tens of thousands of threat reports monthly, cutting mitigation time by more than 90%.

The broader implication is clear: security teams are becoming AI operators, not just incident responders.

The infrastructure race: TPUs vs GPUs heats up

Behind the scenes, the real battle in AI is still infrastructure—and Google is making an aggressive push with its eighth-generation Tensor Processing Units (TPUs).

The new architecture introduces two specialized chips:

• TPU 8t for large-scale training
• TPU 8i for high-efficiency inference

These chips are designed to handle the computational demands of running millions of concurrent AI agents, with improvements in both performance and energy efficiency.

This puts Google in direct competition with NVIDIA, whose GPUs dominate the current AI market. By offering both TPUs and GPU instances, Google is effectively hedging—giving customers flexibility while quietly pushing its own silicon advantage.

For enterprises, the key consideration isn’t just performance—it’s cost per inference at scale. As AI agents multiply, efficiency becomes the deciding factor in ROI.

“Customer zero”: Google tests its own future

One of the more compelling aspects of Google’s strategy is its “customer zero” philosophy—using its own technologies internally before releasing them to the market.

The results provide a rare, real-world benchmark.

AI-generated code becomes the norm

At Google, 75% of new code is now AI-generated and reviewed by engineers, up from 50% just months ago. This isn’t about replacing developers—it’s about amplifying output.

In one case, a complex code migration project was completed six times faster using a combination of engineers and AI agents.

The company’s internal tool, Antigravity, even enabled teams to build a native macOS app prototype in just days—something that traditionally would take weeks.

Security operations at machine speed

AI agents like “CodeMender” are now actively identifying and fixing vulnerabilities in production systems—closing the loop between detection and remediation.

Marketing at scale

For the launch of Gemini in Google Chrome, Google used AI to generate thousands of creative variations, reducing turnaround time by 70% and boosting conversions by 20%.

These examples highlight a key insight: AI’s biggest impact isn’t in isolated tasks, but in compressing entire workflows.

What this means for businesses right now

For organizations watching from the sidelines, Cloud Next ’26 sends a clear message: the barrier to entry is no longer technology—it’s execution.

Here are three practical takeaways:

1. Start with workflows, not models
Instead of asking “Which AI model should we use?”, focus on identifying workflows that can be automated or augmented by agents.

2. Invest in governance early
As agent usage scales, issues like accountability, security, and performance monitoring become critical. Platforms like Gemini’s are built for this phase—not the experimental one.

3. Rethink infrastructure strategy
AI costs don’t scale linearly. Choosing the right mix of compute (TPUs vs GPUs) can significantly impact long-term economics.

Looking ahead to the next phase of AI

As Google prepares for Google I/O in May, the trajectory is becoming clearer.

The industry is moving beyond the novelty of generative AI into something more operational—and more consequential. AI agents are no longer assistants; they are becoming autonomous contributors embedded across the enterprise.

The companies that succeed in this next phase won’t just adopt AI. They’ll learn how to orchestrate it at scale, securely and efficiently.

Cloud Next ’26 suggests Google intends to be the platform where that orchestration happens.