Nvidia Corp. and Accenture plc today are announcing a partnership that will apply 30,000 Accenture consultants trained in Nvidia technology to building artificial intelligence agents for customers across 120 countries.

The newly formed Accenture Nvidia Business Group is aimed at laying the foundation for the next generation of AI functionality using Accenture’s AI platform and a variety of Nvidia technologies. They include Nvidia Inference Microservices, a framework that simplifies the deployment and scaling of AI models. It uses prebuilt, optimized microservices for AI inferencing, and NeMo, a framework for building, fine-tuning, and deploying large-scale AI models for natural language processing, speech recognition and conversational AI.

Generative AI projects drove $3 billion in Accenture revenue in the most recent fiscal year, according to Lan Guan, chief AI officer at Accenture. Nevertheless, the global professional services giant estimated only about 10% of companies have “fully scaled gen AI,” she said.

Agentic AI refers to a type of AI system that can act with some level of autonomy to make decisions and take actions independently based on their environment, goals and training. Whereas traditional AI operates strictly within the bounds of pre-programmed instructions or responses, agentic AI can adapt, learn and respond to situations dynamically.

“It’s no longer just about prompting the pre-built large language models and waiting for response; we can actually develop specialized capabilities that can independently and autonomously interact or actually make progress against goals or human’s intention,” Guan said.

Binders full of agents

Accenture defines several categories of agents. For example, research agents investigate topics and summarize findings to inform decision-making without human intervention. Orchestrator agents are project managers that manage multiple agentic workflow functions.

Using NIMS on Nvidia infrastructure enables models to run between two and five times faster, said Justin Boitano, vice president of enterprise AI software at Nvidia.

“As the AI applications run and interact with people, they generate new data that has to be fed back to improve the models in a continuous learning cycle,” he said. “The Nvidia Nemo framework enables this flywheel through performance-optimized fine-tuning and training. Nvidia AI Foundry is the factory for running this flywheel.”

Accenture’s AI Refinery can be used to build agentic AI systems and customize agents. It includes model training and tuning recipes that enterprises can use to build a models using their own agents. A third component provides access to multiple foundation models and integrates both external and internal data.

The Refinery “harnesses the performance-optimized models of NIMS, uses NIMO for fine-tuning and evaluation and the experts at Accenture are the powerhouse that scales this globally,” Boitano said.

Accenture said the venture’s focus will be on enterprise reinvention, AI-powered simulation and sovereign AI, which transforms functions within an organization. It showed a demonstration of how AI is used in the company’s own marketing communications function to plan campaigns taking into account competitive intelligence, other events, digital assets and historical data from back-end systems.

Image: SiliconANGLE/DALL-E