Databricks and Nvidia: A Strategic Alliance for Agentic AI

Strengthened Collaboration for Agentic AI

In a rapidly evolving artificial intelligence landscape, Databricks and Nvidia have decided to enhance their partnership to focus on the development of agentic AI in professional environments. This collaboration aims to combine accelerated computing, new hardware architectures, and integrated development tools to meet the growing needs of businesses. Companies are no longer just exploring generative AI; they are looking to industrialize agents capable of reasoning and interacting with their business data in secure and regulated environments.

At the recent Data + AI Summit, Databricks unveiled several innovations designed in partnership with Nvidia to accelerate the creation and deployment of AI agents. In addition to GPUs, which are already widely used for training and inference, the two companies are working on a comprehensive infrastructure that will include future NVIDIA Vera processors. These tools are specifically designed for agentic development and aim to provide businesses with the technological foundations necessary to build more autonomous and efficient agents capable of effectively leveraging the organization’s strategic data.

A Common Platform for Enterprise AI

For several years, Databricks has utilized Nvidia technologies to support some of the most demanding AI workloads. Today, this partnership takes a new step with a deeper integration of Nvidia's hardware and software infrastructures within the Databricks platform.

At the heart of this strategy is the Databricks AI Runtime (AIR), an environment that allows data and AI teams to access Nvidia's accelerated computing capabilities without having to manage a complex GPU infrastructure themselves. This means that companies can train, fine-tune, and deploy AI models directly near their governed data. Nvidia Hopper GPUs, combined with the Quantum InfiniBand network, are used for large-scale distributed training, while compatibility with the Blackwell architecture ensures access to future generations of computing power.

Databricks is also democratizing this access by announcing support for Nvidia GPUs in its free edition. This decision could accelerate the adoption of advanced AI by startups, independent developers, and research teams with limited budgets.

Another notable development is the upcoming support for NVIDIA NGC containers and custom CUDA environments. This will allow companies to run their specialized environments directly within the Databricks platform, eliminating the need for multiple layers of external infrastructure. This approach aims to reduce operational complexity while maintaining high performance, a major challenge in AI projects.

New Needs in Computing and Orchestration

The announcement of this collaboration highlights the evolution of AI architectures. Until now, large language models have primarily focused on GPUs. However, future autonomous agents require much more than simple inference capabilities. They must be able to execute tool calls, query databases, and coordinate multiple reasoning steps. They also need to interact with various business systems in real time, making processors essential in this value chain.

To meet this challenge, Nvidia is promoting Vera, its future processor specifically designed for agentic workloads. Unlike traditional CPUs, Vera is optimized for latency-sensitive tasks characteristic of AI agents. According to Nvidia, this architecture could deliver agent performance up to 80% higher and execute SQL queries up to three times faster. The goal is to intelligently distribute processing: Rubin GPUs would handle model inference, while Vera CPUs would manage orchestration, tool calls, and analytical processing.

For businesses, AI agents promise to automate entire business processes. However, their effectiveness will directly depend on how quickly they can access data, interact with applications, and execute their reasoning chains. Any excessive latency then becomes a barrier to adoption. Therefore, the partnership between Databricks and Nvidia aims to anticipate this future issue before it becomes a large-scale problem.

Targeting AI-Intensive Sectors

Databricks and Nvidia are also looking to simplify the development and deployment of agentic AI applications. Databricks Apps will now be able to host the NVIDIA Agent Toolkit, Nvidia's open-source platform dedicated to creating intelligent agents. This will provide companies with an integrated environment to develop, deploy, and govern their agents without leaving the Databricks ecosystem.

This integration brings several essential elements to advanced AI projects, such as multi-step reasoning, the use of external tools, retrieval-augmented generation (RAG), and safeguard mechanisms. Developers will also benefit from Genie Code, an assistant designed to simplify the use of Nvidia infrastructures. This tool will help identify GPU bottlenecks and optimize performance, as well as diagnose CUDA issues in a conversational manner.

However, the ambition of both groups goes beyond generic AI applications. Nvidia is gradually bringing its specialized libraries for various industrial sectors to Databricks. Companies in the healthcare sector will be able to leverage MONAI for medical image analysis or BioNeMo for drug discovery. Players in genomics will benefit from Parabricks, while manufacturers can rely on Omniverse and Isaac Sim for digital twins and robotic simulation.

For Databricks, integrating Nvidia technologies allows it to become a unique platform that combines data, governance, accelerated computing, and development tools. For Nvidia, it is a way to position its infrastructure as the technical backbone of the future agent economy.