Enterprise AI is entering the new adoption phase. The latest wave of generative AI brought focus on content creation and conversational interfaces; the next wave is all about action.

Enterprise AI is entering the new adoption phase. The latest wave of generative AI brought focus on content creation and conversational interfaces; the next wave is all about action. Organizations now expect systems that can reason, decide, and act across workflows, not just respond to prompts.

To bridge AI technologies to business outcomes, NVIDIA and HPE Services are collaborating. HPE is launching an agents hub to help ensure structured and scalable enterprise adoption for agentic AI, generally available today. To help ensure the latest and effective AI capabilities adoption, we are bringing NVIDIA Nemotron™ and HPE’s agents hub together. NVIDIA Nemotron is a family of open models with open weights, training data, and recipes, delivering leading efficiency and accuracy for building specialized AI agents. It is one of the advanced reasoning models we are leveraging to build next-generation agentic applications. Its reasoning and tool-calling capabilities make it well-suited for enterprise agents operating at scale. Within HPE’s agent hub, we are developing and validating agents powered by NVIDIA Nemotron as part of a broader portfolio of reference architectures and reusable patterns.

The hub is a structured foundation that turns these capabilities into production systems. Developed by HPE AI experts with cross-industry experience, it provides the frameworks needed to design, scale, and govern agentic AI. Model-agnostic by design, it helps organizations adopt and operationalize intelligent agents aligned with measurable business outcomes.

NVIDIA Nemotron: A reasoning engine built for agents

NVIDIA Nemotron is engineered for advanced reasoning and collaborative agent systems. Unlike conversational models optimized primarily for dialogue, Nemotron supports structured reasoning, tool invocation, and long-context understanding in high-volume enterprise environments.

NVIDIA Nemotron stands out by taking an unusually open approach for a foundation-model family: NVIDIA is releasing the model weights alongside the data and training recipes used to build it, so developers can inspect, reproduce, and extend the work end to end. That level of transparency helps teams build agentic AI with fewer opaque systems than is typical in today’s foundation-model landscape.

Key technical highlights

• Latent mixture-of-experts (LatentMoE) architecture that interleaves Mamba 2 and MoE layers with selected attention mechanisms to deliver efficient reasoning.
• Super variant adds multitoken prediction (MTP) layers for faster, higher-quality generation.
• NVFP4 quantization to maximize compute efficiency.
• 12 billion active parameters (120 billion total parameters) to balance performance and scalability.
• Support for long context windows up to one million tokens, suitable for use cases with extensive documentation and multistep workflows.

NVIDIA Nemotron is flexible and designed to support a range of enterprise agent applications. Its architecture and reasoning capabilities make it suitable for collaborative, high-volume workloads across IT, business operations, and knowledge-intensive workflows, enabling agents to perform multi-step reasoning, tool integration, and long-context processing.

Nemotron in action: Agentic Knowledge Assistant Suite

HPE recently developed the Agentic Knowledge Assistant Suite, an AI agent within the hub that combines multiple capabilities:

• DocBot + ChatOps: Retrieval-augmented generation over structured and unstructured data from preprocessed input documents
• DocGen: Document generation from templates and information provided by the user
• Document preprocessing: Handling high amounts of complex documents to convert them into tabular insights

Figure 1. Architecture overview with NVIDIA

Structured agentic AI adoption

Adopting agentic AI raises practical questions: which approaches and architectures work; which models are production-ready; how do you govern scale? HPE’s agents hub addresses these questions by providing a curated catalog of validated agents, proven patterns, and lessons learned from real implementations. It helps enterprises reduce risk, accelerate adoption, and make informed decisions about which agentic solutions to deploy in their workflows.

Agents are described using the following 3C framework:

• Characteristics: These define an agent’s operational behaviour and capabilities, such as memory, orchestration, autonomy, perception, and reasoning.
• Categories: Business-focused agent types and maturity levels. Types reflect the role an agent plays in improving efficiency, automation, or decision-making within workflows. Maturity levels indicate how developed or production-ready an agent is, ranging from emergent to fully robust solutions.
• Complexity: Task scope and architectural reach. From task-specific agents to multiagent or mesh architectures.

Organizations leveraging the agent hub benefit throughout their AI adoption journey. Engagement typically begins with feasibility assessment and use case prioritization, followed by validation and prototyping using reference agents. Proof of value projects enable rapid experimentation without starting from scratch, and as solutions mature, enterprises can adopt best practices, governance guidance, and operational lessons embedded in the catalog to scale agentic AI safely and efficiently.

HPE AI Services – Solutions Adoption

HPE’s agents hub and NVIDIA Nemotron are core capabilities within the portfolio of HPE AI Services – Solutions Adoption, empowering AI use cases development and integration in enterprise core processes, and addressing any enterprise AI adoption maturity level:

• Generative AI Discovery explores paths to enable an agentic AI enterprise, guides organizations in identifying and documenting expected outcomes, data, and technology platforms needed to accelerate the implementation of priority AI use cases.
• AI Transformation Workshop delivers a full advisory day focused on prioritizing use cases, considering strategic and tactical approaches, as well as technical requirements to accomplish objectives. Teams participating are from both the technical and business side and it represents a great opportunity to level everyone up to a common understanding of agentic AI vision, ambition, and roadmap to success.
• AI Proof of Value reduces the time to outcomes, moving fast forward from ideation to prototyping AI use cases, gaining stakeholders’ consensus on the impact agentic AI generates for their organization. Key business metrics are tracked and matched with technical performance indicators, all balanced in an agile engagement with an average duration that goes from eight to 12 weeks.
• Generative AI Implementation helps ensure that the selected AI use cases mature into enterprise applications, delivering the MVP ready to onboard final users and perform tasks that produce direct business value. With common engagement duration between three and six months, depending on the complexity of the demand, often dictated by the inferencing-only or fine-tuning need, a crew of AI engineers is dedicated to accelerating agentic AI in the enterprise.

People skills are equally important

AI initiatives will not be successful without addressing people skills. Is your team prepared with the right knowledge? HPE is the only accredited NVIDIA Education Learning Partner offering targeted training for the codeveloped HPE Private Cloud AI solutions from HPE and NVIDIA. HPE provides role-based courses and hands-on labs designed to accelerate operational readiness—from platform operators and AI engineers to business stakeholders responsible for governance and change management. These training programs complement engagements by ensuring teams can deploy, operate, and govern private-cloud agentic AI solutions with confidence.

Pulling it together

In summary, leveraging advanced reasoning models such as NVIDIA Nemotron within structured frameworks such as HPE’s agents hub enables organizations to harness the full potential of agentic AI. As part of the joint NVIDIA and HPE AI portfolio, NVIDIA AI Computing by HPE, these capabilities provide the infrastructure, models, and services organizations need to confidently adopt, operationalize, and scale intelligent, decision-making systems that drive tangible operational and strategic outcomes. As enterprise AI continues to evolve, these integrated solutions pave the way for more autonomous, efficient, and impactful workflows.

Meet the authors:

Raffaele Tarantino, WW AI & Data GTM Strategy Lead, HPE AI Services
Raffaele Tarantino is the WW AI & data GTM strategy lead for the HPE Advisory & Professional Services. He is responsible for the messaging and sales enablement of the services portfolio. Raffaele has 10 years of experience in HPE roles ranging from private cloud consultant to compute specialist and AI architect as a member of the worldwide practice. Raffaele designed the HPE Machine Learning Development Services and contributed to the HPE AI Services – Generative AI Implementation launch.
Contact Raffaele on LinkedIn: Raffaele Tarantino

Ismael Delgado, WW AI Solutions Architect, HPE AI Services
Ismael Delgado is an HPE AI Solutions architect in Advisory & Professional Services with 4 years of experience delivering AI solutions in customer-facing engagements. He specializes in generative AI, AI inference, and agent-based systems, focusing on the technical architecture and design of scalable, production-ready implementations. Ismael has contributed to the creation of the agents hub, developing agent use cases, testing frameworks, and documenting best practices for agent-based solutions.
Contact Ismael on LinkedIn: Ismael Delgado