Google Cloud announces AI investments in Brazil and begins local processing of Gemini Enterprise data

At the Google Cloud Summit Brazil in São Paulo on Thursday, October 24, Google Cloud unveiled a series of initiatives aimed at accelerating the adoption of agentic artificial intelligence in the country. The company plans to double its technical infrastructure in Brazil by 2030 and will start processing Gemini Enterprise customer data locally from October 15 for those using the Gemini 3.5 Flash model.

In addition, the company introduced new tools to help move AI agents from prototype to production, as well as enhanced security, governance, and protection features against attacks on AI models.

A study commissioned by Google Cloud and conducted by consulting firms IDC and Provokers found that 62% of Brazilian organizations already implement or accelerate the adoption of AI agents—the highest rate in the region. However, only 17% of them have managed to consolidate this adoption with the level of governance required to apply it across multiple critical processes.

‘As companies increasingly implement agentic AI, customers tell us they often face the same three challenges: cost, clarity about business value, and the need for enterprise-level controls,’ said Thomas Kurian, CEO of Google Cloud.

Data processing within Brazil

From October 15, the web version of Gemini Enterprise will process customer data locally in Brazil for those using Gemini 3.5 Flash. This move complements the existing local storage of agent workflow data and meets the country’s regulatory and data sovereignty requirements.

To support this operation, virtual machines equipped with Nvidia RTX PRO 6000 Blackwell Server Edition GPUs are now available in the São Paulo cloud region, designed for heavy and real-time inference workloads.

Tools to move agents from prototype to production

Google Cloud also launched new features to reduce friction between the testing phase and the deployment of AI agents in corporate environments. These new features include:

• Projects: Dedicated spaces within Gemini Enterprise, linked to specific files, that maintain history and context over time for individual or team use.
• Skills Registry: A central registry of “skills” (sets of instructions and resources) that teach an agent to perform specific tasks, with usage policies defined by agent, person, or team.
• Agent Sandbox: An isolated and managed environment where agents can compile and execute code, use the command line, or operate a full computing environment, including web browsing.
• AlloyDB for Agents: A PostgreSQL-based database designed to support millions of AI agents operating simultaneously without impacting a company’s production systems.

Security tailored to the speed of agents

In partnership with cybersecurity firm Wiz, Google Cloud will offer Brazilian companies a free risk exposure assessment using the AI-based penetration testing tool Red Agent. The solution analyzes the external attack surface and generates a prioritized vulnerability report.

The company has also expanded Model Armor, its system for protecting against attacks on AI models. The new version includes image screening, enhanced detection of attempts to manipulate commands through Google Workspace files, a context window of up to 64,000 tokens, and support for Anthropic models hosted on Gemini Enterprise.

Another new feature is Auth Manager, which eliminates the need for custom code to manage credentials and authentication in automated workflows between agents.

Google Cloud reiterates that, in any region, customer data and results are not used to train the company’s language models without explicit consent.