NVIDIA Nemotron and Accenture AI Refinery: Building Reasoning AI Agents for Enterprise

Updated: May 06 2025 05:36

AI Summary: NVIDIA and Accenture are accelerating the shift to reasoning and agentic AI, moving beyond basic language models. NVIDIA's new Llama Nemotron models offer enhanced reasoning, coding, and decision-making capabilities with significant performance gains, forming a foundation for advanced AI agents. Concurrently, Accenture's expanded AI Refinery platform, powered by NVIDIA, introduces a no-code AI agent builder, empowering business users to create and customize agents easily.

We're moving beyond simple language models to what industry leaders are calling "reasoning AI" and "agentic AI platforms." The recent announcements from NVIDIA and Accenture highlight just how quickly this technology is advancing and what it means for businesses across all sectors.


On March 18, 2025, NVIDIA unveiled its open Llama Nemotron family of models with enhanced reasoning capabilities. These models aren't just incremental improvements—they represent a fundamental shift in how AI can support business operations.

Built on Llama models, NVIDIA has enhanced these reasoning models during post-training to significantly improve capabilities in:

• Multistep mathematical operations
• Complex coding tasks
• Logical reasoning
• Advanced decision-making

The results are impressive: up to 20% improved accuracy compared to base models and 5x faster inference speed compared to other leading open reasoning models. This means enterprises can now handle more complex reasoning tasks, enhance decision-making capabilities, and reduce operational costs.


NVIDIA is offering these models in three distinct sizes, each optimized for different deployment scenarios:

• Nano: Delivers the highest accuracy for PCs and edge devices
• Super: Offers the best balance of accuracy and throughput on a single GPU
• Ultra: Provides maximum agentic accuracy for multi-GPU server deployments

This tiered approach ensures organizations of all sizes and technical capabilities can leverage these reasoning models according to their specific needs and infrastructure constraints.

The adoption rate speaks volumes about the potential impact of these models. Industry leaders already working with NVIDIA's Llama Nemotron reasoning models include:

• Microsoft (integrating the models into Azure AI Foundry)
• SAP (enhancing their Business AI solutions and Joule AI copilot)
• ServiceNow (building more accurate AI agents)
• Accenture (incorporating the models into their AI Refinery platform)
• Deloitte (planning integration into their Zora AI agentic platform)

Walter Sun, global head of AI at SAP, emphasized how these models will "refine and rewrite user queries, enabling our AI to better understand inquiries and deliver smarter, more efficient AI-powered experiences that drive business innovation."


On the same day, Accenture announced significant expansions to its AI Refinery platform, which is built on NVIDIA AI Enterprise. The centerpiece of this announcement is a new AI agent builder that democratizes the creation of AI agents, allowing business users to build and customize agent teams without coding expertise.


This represents a paradigm shift in how businesses can deploy AI. Instead of relying on technical teams, business decision-makers can now directly modify agents in response to:

• Policy changes
• New products or competitive offers
• Customer feedback
• Demand fluctuations

According to Lan Guan, chief AI officer at Accenture, this capability empowers "organizations and decision-makers with the flexibility to drive business value faster, with agents that can observe the environment, apply reason, continuously improve and take action."


Accenture is already implementing AI Refinery solutions with several major organizations:

• ESPN: Testing an AI-powered avatar called FACTS to revolutionize sports fan experiences
• HPE: Developing agentic AI solutions for category and sourcing strategies, spend management, and contract obligation management
• Noli: Powering an AI-driven beauty engine that delivers hyper-personalized routines and product recommendations
• United Nations: Creating a multilingual research agent supporting over 150 languages to promote awareness of Sustainable Development Goals

Perhaps most impressive is Accenture's aggressive push to develop industry-specific AI agent solutions. They're currently working on more than 50 solutions, with a goal of reaching 100 by year-end. These solutions span telecommunications, financial services, insurance, manufacturing, healthcare, retail, and more.

Some examples already available include:

• Agent assist for telecommunications: Empowering call center professionals with real-time insights, resulting in 25X faster call processing and 24% improvement in call accuracy
• Insurance underwriting: Automating property and casualty underwriting tasks, enabling insurers to process 100% of coverage submissions instead of leaving up to 50% untouched
• Order-to-cash: Streamlining finance operations by automating order validation, invoice reconciliation, and accounts receivable management
• Commercial credit sales intelligence: Transforming commercial banking by providing immediate, personalized support to credit underwriters

To further support developers, NVIDIA is releasing several new building blocks as part of the NVIDIA AI Enterprise software platform:

• NVIDIA AI-Q Blueprint: Enables enterprises to connect knowledge to AI agents that can autonomously perceive, reason and act
• NVIDIA AI Data Platform: A customizable reference design for enterprise infrastructure with AI query agents
• New NVIDIA NIM microservices: Optimize inference for complex agentic AI applications
• NVIDIA NeMo microservices: Provide solutions to establish and maintain robust data flywheels that enable continuous learning

Traditional language models excel at generating text and following instructions, but they often struggle with complex reasoning tasks that require multiple steps of logical thinking. NVIDIA's new models address this limitation through extensive post-training on:

• High-quality curated synthetic data generated by NVIDIA Nemotron and other open models
• Additional curated datasets co-created by NVIDIA

What's particularly important is NVIDIA's commitment to openness—the tools, datasets, and post-training optimization techniques used to develop these models will be openly available. This gives enterprises the flexibility to build their own custom reasoning models tailored to their specific needs.

Accenture's research validates the significant impact of these technologies. Organizations that have scaled at least one industry-tailored solution for a core process are three times more likely to have delivered ROI exceeding expectations. These reasoning AI agents can:

• Reduce operational costs through automation of complex cognitive tasks
• Improve decision-making with more accurate and comprehensive analysis
• Accelerate response times to market changes and customer needs
• Enable non-technical business users to directly shape AI capabilities
• Create competitive advantages through industry-specific optimizations

As these technologies become more widely available and adopted, we can expect several developments:

1. Increased collaboration between AI agents and human workers, with AI handling routine cognitive tasks while humans focus on creativity and judgment
2. More sophisticated agent networks that can coordinate across different business functions
3. Greater customization of AI capabilities for specific industry needs and challenges
4. Accelerated innovation cycles as businesses can more quickly implement and iterate on AI solutions

The announcements from NVIDIA and Accenture mark a pivotal moment in the evolution of enterprise AI. We're moving beyond simple language models to truly reasoning agents capable of handling complex, multi-step challenges across diverse business domains. As Jensen Huang, founder and CEO of NVIDIA, noted, "Reasoning and agentic AI adoption is incredible." With these new tools, enterprises everywhere now have "the building blocks to create an accelerated agentic AI workforce."

NVIDIA Llama Nemotron Nano and Super models: https://build.nvidia.com
NVIDIA AI-Q Blueprint: https://blogs.nvidia.com/blog/ai-agents-blueprint/
NVIDIA NIM microservices:
Hugging Face: https://huggingface.co/collections/nvidia/llama-nemotron-67d92346030a2691293f200b
Github: NVIDIA Agent Intelligence toolkit