Google Cloud adds OpenAI models with private data protections

Expanding the AI toolkit

Google Cloud is expanding its offerings by integrating OpenAI models into its cloud infrastructure. This initiative begins with a private preview specifically designed for selected business customers who need diverse AI options. The launch includes several specific models, including GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna.

The company is positioning these models to serve different types of professional needs. For instance, some models are intended for demanding work that requires high reasoning capabilities, while others are meant for everyday tasks that need fast responses. By offering these choices, Google aims to provide a versatile toolkit for different business functions.

This move represents a shift toward a multi-model environment where users are not locked into a single provider. Customers can evaluate how different models perform on specific tasks to find the right balance of speed and capability. This flexibility is intended to help companies optimize their AI workflows based on actual performance results.

Simplified billing and access

One of the primary goals of this integration is to streamline how businesses purchase and manage AI services. Companies can acquire access to OpenAI models through the Google Cloud Marketplace. This means that all usage will appear on a single Google Cloud invoice rather than requiring separate billing systems for different providers.

The integration also allows these costs to count toward existing cloud spending commitments. Specifically, the models will qualify for Flexible Savings Plans, which provide lower costs for tokens—the basic units of AI processing—in exchange for a monthly spending commitment. This structure helps large organizations manage their budgets more predictably across different AI services.

Michael Gerstenhaber, VP of Product Management for Gemini at Google Cloud, noted that customers can switch models based on performance, response times, and cost. This administrative simplicity is designed to reduce the friction of adopting new technologies. It allows IT departments to manage a complex array of AI tools through a familiar and centralized procurement process.

Enterprise security standards

A major hurdle for many large companies is the fear that their private data will be used to train public AI models. Google addresses this by promising that customer prompts, context, and outputs will remain within designated data boundaries. The company explicitly states that this data will not be used to train its foundation models.

Yashodha Bhavnani, VP of AI products at Box, highlighted that this setup allows companies to use frontier models while maintaining rigorous compliance standards. By keeping the data within Google Cloud's security perimeter, businesses can leverage OpenAI's capabilities without compromising their internal information. This approach is intended to help organizations operate more efficiently with their enterprise content.

Beyond basic privacy, Google plans to introduce even stricter protections for these models. These include zero data retention options, where data is not stored after processing, and Confidential Compute protections. These features are designed to meet the high security requirements of industries that handle sensitive personal or corporate information.

Developer and workflow integration

The new preview offers several technical pathways for developers to integrate these models into their existing workflows. Customers with preview access can build new applications using the models directly through Google Cloud's APIs. This allows for the creation of custom tools that can be deployed across an entire organization.

Developers can also connect their local Codex CLI and Codex Desktop workflows to Google’s Gemini platform for inference. In this setup, Google Cloud handles the processing of the model requests while the developer maintains their local workflow. This hybrid approach allows for more seamless integration into existing coding and development environments.

Looking ahead, Google intends to bring OpenAI models to Gemini for business and Google Antigravity. These integrations will further unify the user experience. By providing a single point of access for multiple models, Google aims to make it easier for companies to deploy AI at scale across various departments.

Evaluating performance and cost

The private preview serves as a testing ground for companies to determine which models provide the best return on investment. Because different models have different price points and speeds, businesses must assess the specific needs of their use cases. For example, a customer-service assistant might use a faster, cheaper model to sort basic messages.

Conversely, that same assistant might switch to a more capable and expensive model to investigate complex customer complaints. This tiered approach allows companies to save money on high-volume, simple tasks while reserving high-powered AI for complex reasoning. The goal is to find the optimal point where the results justify the ongoing costs of the model.

As companies move forward, they will need to establish internal policies regarding data access and activity recording. They must decide who within the company can access specific information and what happens to data after a request is processed. These governance steps are essential for any organization looking to put AI to work across its enterprise content safely.