NVIDIA AI Factory

NVIDIA AI Factory is a full-stack infrastructure that manages the entire artificial intelligence lifecycle—from data collection and preparation to model deployment. It helps organizations quickly and reliably build production environments for AI.

There are two deployment concepts and options. The On-Premises AI Factory follows the standard approach of deploying your own hardware and infrastructure directly in the organization’s data center for maximum control, data security, and independence. In contrast, the Cloud AI Factory shifts computing power to cloud service providers, enabling lightning-fast deployment of AI workloads with immediate scalability and flexibility.

Exascale

Up to 3.6 exaFLOPs per rack

Modular

Scalable architecture

Software

NVIDIA AI Enterprise

Liquid Cooling

100% water-cooled

Services

Full NVIDIA Support

Design

Reference Architecture

* Specifications may vary depending on the configuration.

Benefits

Deployment Speed

The infrastructure uses pre-configured rack-level designs and an integrated software stack, which simplifies the implementation process and reduces the time required to bring the system online.

Performance and Efficiency

The architecture is primarily optimized for the “tokens per watt” metric—more useful AI work per watt, along with lower operating costs while delivering higher performance.

Verified Ecosystem

This is a full-stack solution built on NVIDIA accelerated computing. The entire system has been tested and certified within a partner ecosystem that includes hardware manufacturers (Dell, HPE, Lenovo, Supermicro, Cisco), ensuring standardized compatibility.

Scalability

Computing capacity can be flexibly scaled from an initial few nodes up to clusters containing thousands of GPUs, depending on current system requirements.

What is an AI Factory?

An AI factory is a specialized infrastructure designed to “produce” intelligence—tokens, predictions, and decisions—on a large scale. Unlike a traditional data center, the entire stack (accelerated hardware, network, storage, and software) is built and optimized around a single goal: maximum AI throughput at the lowest cost per token and per watt. It covers the entire AI lifecycle—from data preparation through training and fine-tuning to inference and agent-based applications.

It is precisely the transition to the era of AI reasoning and agent-based AI that is changing computing requirements: post-training may require up to 30 times more computing power than traditional pre-training, and so-called test-time scaling may require up to 100 times more than standard inference. That is why AI infrastructure is designed like a factory—with performance, energy efficiency, and scalability in mind.

Hardware Software NVIDIA NGC Support

Hardware Foundation

NVIDIA AI Factory is not limited to a single specific server type or fixed hardware configuration. Its computing infrastructure consists of systems from the NVIDIA Validated Design program, which ensures maximum flexibility while maintaining fully guaranteed performance and reliability.

Whether it’s cutting-edge reference solutions from the NVIDIA DGX family or certified servers from leading OEM partners, the infrastructure can be built in a modular fashion using the most powerful architectures. Computing capacity thus seamlessly covers needs ranging from systems equipped with Hopper and Blackwell accelerators to the next generation of gigascale AI platforms.

Thanks to standardized hardware building blocks at the rack level, physical deployment is fast and linearly scalable. Furthermore, in the latest exascale solutions, such as the massive NVIDIA DGX Vera Rubin NVL72 rack system, full liquid cooling is an integral part of the hardware design. This enables extremely high computing density per rack and radically increases the energy efficiency of the entire infrastructure.

Network Infrastructure

To enable thousands of graphics cards to function as a single gigantic brain, AI Factory utilizes a specially designed network infrastructure. Furthermore, in the emerging era of so-called gigascale AI, these systems are being expanded to meet the needs of interconnecting hundreds of thousands of computing units (GPUs and CPUs). The interconnect layer is built on NVIDIA Quantum InfiniBand (for maximum throughput and low latency during model training) and NVIDIA Spectrum-X (an Ethernet solution optimized specifically for AI workloads).

The latest generation of the network features the NVIDIA Spectrum-6 Ethernet switch with a throughput of 102.4 Tb/s, which offers twice the capacity of previous systems and is part of the new NVIDIA Vera Rubin platform. Compared to standard off-the-shelf Ethernet, Spectrum-X technology delivers up to 1.6 times higher network performance for AI and can maintain 95% network efficiency even in deployments exceeding 100,000 GPUs. The new hardware infrastructure also fully supports liquid cooling, enabling a comprehensive cooling approach and increasing the network’s energy efficiency.

The entire system is designed modularly using scalable units (SUs), allowing for seamless expansion of the infrastructure to precisely meet your needs without disrupting existing operations.

Reference Architectures

NVIDIA
RTX PRO AI

An architecture designed for the enterprise sector. It offers the ideal solution for industrial artificial intelligence, visual computing, and reliable inference.

Read Whitepaper

NVIDIA
nd HGX AI

Scalable architecture for demanding data centers. It focuses on large-scale LLM training, advanced generative AI, and complex fine-tuning.

Read Whitepaper

NVIDIA
Vera Rubin

A next-generation platform ready for the era of gigascale AI. A system designed for maximum performance and efficiency when training models with trillions of parameters.

Read Whitepaper

Storage

Of course, the entire AI Factory also requires high-performance storage. That’s why the systems are paired with storage solutions from the world’s leading providers (VAST, DDN, or Everpure). This integration creates a comprehensive and fully optimized system for AI training and inference, ensuring you’ll always have ample storage capacity.

VAST

VAST offers unified storage with NVMe and QLC flash, optimized for AI, HPC, and big data. It utilizes advanced compression and deduplication.

More information

DDN

DDN specializes in high-performance storage for HPC, AI, and enterprise applications. It offers scalable solutions that support parallel access to data.

More information

PURE

Everpure (formerly Pure Storage) provides all-flash storage with a focus on simplicity, performance, and low latency. It is popular for enterprise environments and cloud services.

More information

DGX SuperPOD – AI Factory

NVIDIA DGX SuperPOD offers a comprehensive AI data center as a solution for companies building AI Factories, seamlessly providing cutting-edge computing power, software tools, expertise, 24/7 support, and continuous innovation. With a choice of computing options, it enables every enterprise to integrate AI into its business and create innovative applications without having to grapple with the complexity of the platform.

Software

Of course, the NVIDIA AI Factory infrastructure is supported by an extensive software ecosystem that handles everything from management to automation and simplifies the deployment of AI and HPC applications. Everything is unified under the NVIDIA Base Command Manager platform, which greatly simplifies the management and control of the entire system from a single location. Given that the AI Factory can scale from dozens to more than a thousand GPUs, this management approach can save a tremendous amount of time.

At the same time, the NVIDIA AI Enterprise software suite helps you deploy and train AI models by automating time-consuming tasks—from preparing training data to performing inference on the finished model. The path to production-ready AI is accelerated primarily by NVIDIA NIM™ microservices and NVIDIA NeMo™ tools for model customization and data processing.

NVIDIA AI Enterprise

NVIDIA AI Enterprise is a cloud-native software platform that simplifies the development and deployment of production-ready generative AI applications. It provides a suite of tools, such as NVIDIA NIM™ microservices for efficient inference and NVIDIA NeMo™ for data processing and model fine-tuning.

The platform supports deployment across a variety of environments—from the cloud to data centers to edge devices and workstations. NVIDIA AI Enterprise ensures enterprise-grade security, stability, and technical support, making it easier to move from prototype to production.

More About AI Enterprise

NVIDIA Base Command Manager

NVIDIA Base Command Manager is a comprehensive software solution for orchestrating and managing AI and HPC clusters. It enables rapid deployment and end-to-end management of various clusters in data centers, at the network edge, and in multi- and hybrid cloud environments. It automates node provisioning and administration across the entire AI Factory infrastructure and enables orchestration with Kubernetes.

It integrates with NVIDIA platforms, including DGX systems, BlueField DPUs, and NVIDIA Spectrum switches. Base Command Manager provides tools for infrastructure monitoring and job management, and ensures enterprise-level security and stability.

More about Base Command Manager

NVIDIA GPU Cloud Catalog

The NVIDIA NGC catalog is an extensive software repository that contains GPU-optimized containers, pre-trained AI models, Helm charts, and industry-specific microservices for developing machine learning and deep learning applications, HPC, or visualizations. Deploying these applications is as simple as copying the link to the relevant container and running it on a certified system. Individual development environments—including versions of all included libraries, frameworks, and parameter settings—are continuously updated and optimized by NVIDIA to achieve maximum performance across the entire NVIDIA AI Factory architecture.

NVIDIA NGC Catalog

Key Benefits of NVIDIA Support

The solution’s key strength the is support for the entire ecosystem. Hardware support (in the event of a component failure) is a given. Software support for the entire environment is essential in case something isn’t working as it should. Customers have access to hundreds of developers ready to help. Support is an integral part of the NVIDIA AI Factory solution. It is available for 3–5 years and can be extended beyond that period.

Repair by a technician at the customer’s site

You get professional service right at your doorstep. A technician will arrive on-site, perform a diagnosis, and make the repair quickly and efficiently.

Remote diagnostics

Quick remote assistance. Our specialists will connect to your device and perform diagnostics and repairs without the need for a personal visit.

Extended software support

Provides longer access to updates, bug fixes and software enhancements. Your device stays up-to-date and protected even after warranty.

Latest updates and upgrades

Get access to the latest software and firmware releases that improve performance, security and add new device features.

Direct contact to NVIDIA support

Don’t hesitate to communicate with the NVIDIA team of experts. Quick answers, technical assistance and advice tailored to your needs.

Private NGC Storage and Software

Secure and private storage for your software and data. Quick access to download tools and documentation from the NVIDIA environment.

Installation

The entire NVIDIA AI Factory solution is delivered on a “turnkey” basis—you, the customer, will receive a fully functional and connected system ready for immediate use. The entire installation is carried out by specially trained technicians using specialized equipment for deploying AI infrastructure. The data center then receives full NVIDIA support, including a dedicated support specialist for your system.

NVIDIA Price Promotions

NVIDIA offers special pricing promotions and programs for the education sector (EDU) and startups on DGX systems. We’ll provide you with a quote.

Get a quote

M Computers represents NVIDIA on the Czech market in the enterprise sector of computing accelerators and AI systems.

We are the first company in Central and Eastern Europe to achieve the highest ELITE PARTNER status, as well as to receive the NVIDIA AI Innovator, NVIDIA AI Champion, and NVIDIA AI Expert designations.

NVIDIA Deep Learning Institute

The NVIDIA Deep Learning Institute (DLI) offers both online and hands-on training for developers, data scientists, and researchers looking to solve challenging problems with deep learning and accelerated computing.

More information

Frequently Asked Questions (FAQ)

What is an AI Factory?

AI Factory is a data center designed to produce intelligence—tokens, predictions, and decisions—on a large scale. Unlike a typical data center, the entire stack (GPUs, network, storage, software) is optimized for AI throughput and the cost-per-token-per-watt metric. It covers the entire AI lifecycle, from data preparation to inference.

What does “tokens per watt” mean, and why does it matter?

Tokens per watt is a metric for AI infrastructure efficiency—it measures how much useful AI work (tokens) you can produce per watt of energy consumed. In continuous AI operations, energy costs are the main operating expense, so higher efficiency directly reduces the cost per token.

What is the difference between On-Premises and Cloud AI Factory?

The On-Premises AI Factory involves deploying hardware and infrastructure directly within an organization’s own data center, ensuring maximum control, independence, and data security. Cloud AI Factory, on the other hand, shifts computing power to cloud service providers, ensuring lightning-fast deployment of workloads and immediate flexibility and scalability.

Can an AI Factory be expanded later?

Yes. Using scalable units (SU), the entire system can be flexibly scaled up from an initial few nodes to clusters containing thousands of GPUs, all without disrupting existing operations.

How long does it take to deploy an AI Factory?

Thanks to NVIDIA’s validated reference architectures and pre-built rack-level designs, deployment is significantly faster than designing from scratch. We deliver turnkey systems —design, integration, installation, and commissioning. The specific timeline depends on the scope and availability.

What is the difference between an AI Factory and a traditional data center?

A traditional data center stores and processes data; the AI Factory produces intelligence. It is built on accelerated computing (GPU), designed for massive parallel workloads, and optimized for energy efficiency (tokens per watt) rather than general computing power.

What hardware is NVIDIA AI Factory built on?

The computing infrastructure is not based on a single specific server, but rather on systems from the NVIDIA Validated Design program. The infrastructure can be built modularly using reference solutions from the NVIDIA DGX family or certified servers from OEM partners such as Dell, HPE, Lenovo, Supermicro, and Cisco. In terms of hardware , it supports systems with Hopper and Blackwell accelerators , as well as the upcoming gigascale generation.

What reference architectures does NVIDIA AI Factory offer?

Customers can choose from three main reference architectures:

NVIDIA RTX PRO AI: An architecture designed for the enterprise, ideal for industrial AI, visual computing, and inference.

NVIDIA HGX AI: A scalable data center solution designed for large-scale LLM training and advanced generative AI.

NVIDIA Vera Rubin: A next-generation gigascale platform designed for training models with trillions of parameters.

What do NVIDIA AI Enterprise and the NGC catalog offer customers?

The NVIDIA AI Enterprise software suite simplifies the development of production-ready AI applications using NVIDIA NIM™ microservices (for inference) and NVIDIA NeMo™ tools (for data processing). The NVIDIA NGC catalog then provides GPU-optimized containers, pre-trained models, and Helm charts that can be deployed simply by copying a link to a certified system.

How is network communication between nodes in AI Factory handled?

The interconnect layer uses NVIDIA Quantum InfiniBand networking technology for maximum throughput during training and NVIDIA Spectrum-X as an Ethernet solution optimized for AI workloads. The latest generation features the NVIDIA Spectrum-6 switch, which delivers 102.4 Tb/s of throughput and maintains 95% network efficiency even when deployed with over 100,000 GPUs.

Non-binding inquiry

Are you interested in the NVIDIA AI Factory infrastructure and would like to deploy it at your organization?
Please feel free to contact us via the contact form, or reach out directly to our NVIDIA product specialist.

Kamila Jeřábková

NVIDIA, AI solutions

+420 734 161 516

kamila.jerabkova@mcomputers.cz