VMware kicked off its Explore event in Las Vegas with several announcements focused on empowering enterprises to develop generative AI solutions.

At the event, VMware and Nvidia revealed an expanded partnership, introducing the VMware Private AI Foundation with Nvidia. This new offering provides businesses with the tools and computing resources needed to fine-tune large language models and deploy AI-powered applications using proprietary data within VMware’s cloud infrastructure.

For many businesses, using public AI models isn’t an option due to concerns about data privacy and the unknown sources of training data. In fact, a recent survey by AI platform Predibase found that over 75% of enterprises are hesitant to use commercial large language models in production because of these privacy concerns.

The solution lies in custom models that are trained using secure, company-specific data. VMware highlights its multi-cloud approach as a secure and flexible way to build tailored AI models that meet these needs.

The VMware Private AI Foundation with Nvidia is a suite of integrated AI tools that allows businesses to deploy AI models trained on private data in data centers, public clouds, or edge environments. This architecture is built on VMware’s Cloud Foundation and integrates with Nvidia’s AI Enterprise software and compute infrastructure.

VMware CEO Raghu Raghuram emphasized that enterprises must be able to maintain data privacy and minimize IP risk when training and deploying AI models. With VMware Private AI, businesses can leverage their trusted data to build and run AI models more securely and quickly in a multi-cloud environment.

Enterprises now have the flexibility to choose where to build and run their AI models, thanks to a secure architecture. VMware and Nvidia claim that AI workloads can scale across up to 16 GPUs in a single virtual machine, with the ability to expand across multiple nodes, which can help reduce costs and improve efficiency. Additionally, VMware’s vSAN Express Storage Architecture offers optimized NVMe storage and supports GPUDirect storage over RDMA, enabling direct I/O transfer from storage to GPUs without involving the CPU.

The new platform also includes Nvidia’s NeMo AI framework, which is part of Nvidia AI Enterprise. NeMo combines tools for model customization, guardrail toolkits, data curation, and pretrained models. It uses TensorRT to optimize inference performance on Nvidia GPUs. VMware and Nvidia say enterprises can use the Nvidia AI Workbench to pull models, like Llama 2 from Hugging Face, customize them remotely, and deploy production-grade generative AI in VMware environments.

Nvidia’s CEO, Jensen Huang, noted that their extended partnership with VMware would provide businesses across industries like financial services, healthcare, and manufacturing with the software and computing power they need to harness the potential of generative AI using custom applications built with their own data.

However, Nvidia is not the only player in the AI space. Many organizations are turning to open-source solutions to have the flexibility to use a variety of tools and frameworks. To address this, VMware introduced the VMware Private AI Reference Architecture for Open Source. This architecture integrates open-source technologies from VMware’s partners to provide an open framework for building and serving models on VMware Cloud Foundation.

A key partnership in this effort is with Anyscale, the creators of Ray, a widely used open-source compute framework. VMware’s Cloud Foundation enables data scientists and machine learning engineers to scale AI and Python workloads with Ray, using existing computing resources rather than relying on public cloud services.

Anyscale’s CEO, Robert Nishihara, explained that companies are under pressure to stay ahead in AI while scaling and iterating quickly. He noted that Ray’s ability to run on any cloud provider, on-premises, or even on a laptop makes it a natural fit for VMware’s customers, who run across diverse environments.

Chris Wolf, VP of VMware AI Labs, highlighted that AI has traditionally been developed by data scientists for data scientists. With the new VMware Private AI offerings, VMware aims to bring AI models and compute resources closer to the data, making it more accessible for a wider range of enterprise users. This approach will benefit many business use cases, from software development and marketing content generation to customer service and extracting insights from legal documents.

In addition to the Private AI offerings, VMware also introduced Intelligent Assist, a suite of generative AI solutions designed to automate aspects of enterprise IT in multi-cloud environments. Intelligent Assist will be integrated into several VMware products, including VMware Tanzu, which will help manage multi-cloud visibility and configuration. Users will be able to make requests and refine changes to cloud infrastructure through conversational prompts. Workspace ONE will also integrate this technology, enabling users to create high-quality scripts using natural language. The new generative AI capabilities will also enhance NSX+, helping security analysts better assess security alerts and respond more effectively to threats.