Oracle to host Google TPUs and OpenAI chips in its cloud

A new hardware ecosystem

Oracle Corp. announced it will make Google's TPU artificial intelligence accelerators available through its public cloud. This partnership represents a significant shift in the industry, as it is the first time Google is supplying its custom chips to another hyperscaler. Oracle plans to begin deploying these chips next year and intends to gradually increase capacity from that point forward.

In addition to Google's hardware, Oracle is expanding its portfolio with other specialized AI silicon. The company will also host Cerebras Systems' wafer-scale accelerators, which are large chips containing trillions of transistors. Furthermore, Oracle is integrating OpenAI's Jalapeño inference processor, a chip that OpenAI claims can run certain open-source models 1.5 times faster than competing silicon.

By offering a variety of chips from different manufacturers, Oracle is positioning its cloud as a versatile hub for AI development. This allows customers to choose the specific hardware that best fits their performance needs. The move highlights a growing trend of infrastructure providers diversifying their hardware options beyond their own internal designs.

How the networking works

To make these chips functional within its data centers, Oracle is utilizing specific networking technologies from Google. The TPUs will exchange data using two methods: ICI and OCS. While ICI transmits data over copper wires, OCS uses optical links to move data between servers.

The OCS technology is designed to be more efficient because it avoids the common step of converting light signals into electricity and back again. Instead, it uses tiny mechanical mirrors to steer light beams directly toward the recipient servers. This method helps lower the overall power usage of the data center while maintaining high speeds.

Oracle will connect these TPU clusters to the rest of its host data center using its own Acceleron networking technology. This system uses a custom SmartNIC, which is a specialized chip designed to handle network management and cybersecurity tasks. According to Oracle, this setup allows for high throughput and low latency when moving data between databases, storage, and the new TPU clusters.

Automating business workflows

Beyond hardware, Oracle is introducing several software integrations to help businesses use AI in their daily operations. A new service called AI Workspace will connect Oracle's cloud applications with ChatGPT. This allows the chatbot to perform tasks on behalf of workers, such as managing customer records or coordinating across different apps.

Oracle describes a scenario where a customer order is at risk due to a supplier delay. In this case, the AI Workspace could find an alternative product, check inventory, and update the order automatically. The system is designed to maintain a specific goal while coordinating the various agents, data, and tools required to finish the job.

To support this, Oracle is launching App Composer, a tool designed for non-technical users to build custom software. Users can use AI assistants to write the code, and the tool then scans the software for vulnerabilities. Oracle will manage the deployment and operations of these apps while the customer maintains control over data and costs.

Security and small business

Security is a major component of Oracle's new AI rollout, specifically through a tool called AI Security Mesh. This service uses OpenAI's cybersecurity-optimized AI models to identify weak points in a customer's applications. It automatically prioritizes these vulnerabilities based on how severe they are and provides a suggested fix for each one found.

For smaller companies, Oracle is launching the AI SMB Bundle, which is a 'free-to-start' product suite. This bundle includes several of the new tools, such as AI Workspace, App Composer, and ChatGPT Business. It also includes the NetApp QuickStart accounting platform and Codex, providing a comprehensive suite of AI-driven business tools.

As reported by SiliconANGLE, these small business customers will also receive $12,000 in cloud credits every year. This initiative is intended to lower the barrier to entry for smaller firms looking to adopt AI. By bundling these services, Oracle aims to provide a streamlined way for small businesses to integrate AI into their existing accounting and workflow systems.

}

The path forward for cloud

The integration of Google's TPUs into Oracle's cloud marks a new era of cooperation between major tech giants. By sharing hardware and software, these companies are creating a more interconnected environment for AI development. This allows enterprises to build on a foundation of diverse technologies rather than being locked into a single provider's ecosystem.

For the end user, this means that the complexity of choosing the right AI infrastructure is being simplified. Oracle is taking on the heavy lifting of managing the hardware, networking, and security. This allows business owners to focus on the outcomes, such as faster order processing or more secure software deployment.

Looking ahead, the success of these partnerships will likely depend on how well the different technologies can communicate. Oracle's use of its own Acceleron networking to bridge Google's chips is a key technical hurdle they are addressing now. As more companies join this ecosystem, we can expect to see more cross-platform integrations and specialized AI hardware becoming standard in the cloud.

}