# NVIDIA AI Factories Enhance Revenue with High Throughput and Durability > Discover how NVIDIA's AI factories are setting a new standard in AI infrastructure, delivering unmatched productivity and cost efficiency while maximizing ROI for operators in an evolving market. **Source**: blogs.nvidia.com | **Published**: 2026-10-01 | **Type**: article ## Key Facts - NVIDIA AI factories yield 30x higher throughput, enhancing revenue potential significantly. - Durable GPUs like A100 retain 25% value after 6 years, extending financial viability for operators. - Fungibility allows NVIDIA systems to handle diverse workloads, ensuring consistent demand and utilization. - Lower token costs expand use cases, driving increased demand for NVIDIA's AI computing solutions. - Continuous software optimization keeps older hardware competitive, reducing replacement costs for users. ## Summary NVIDIA has established a new paradigm in the AI infrastructure market through its innovative AI factories, which are designed to optimize return on investment (ROI) across three critical dimensions: earning capacity, useful life, and demand. Each AI factory, costing approximately $60 million per megawatt, is engineered to deliver maximum productivity, durability, and fungibility, making them a compelling choice for operators seeking to capitalize on the burgeoning AI landscape. The earning capacity of an AI factory is fundamentally linked to its ability to produce tokens—units of value generated from AI workloads. NVIDIA's latest systems, such as the Vera Rubin NVL72, reportedly achieve over 30 times higher throughput per megawatt compared to older models, while also reducing the cost per million tokens by as much as 45%. This efficiency is achieved through extensive engineering codesign, which ensures that every component of the factory—from hardware to software—is optimized for peak performance. As token production becomes cheaper, the demand for computational power expands, contrary to the assumption that lower costs would diminish demand. In fact, cheaper tokens enable a wider array of use cases, driving further consumption of computational resources. Durability is another cornerstone of NVIDIA's AI factories. The longevity of hardware is crucial for maximizing ROI, and NVIDIA’s GPUs, such as the A100, continue to generate revenue long after their initial deployment. For instance, CoreWeave has extended bookings for A100 units from 2020 through 2029, demonstrating the sustained economic viability of older systems. Analysts note that the depreciation schedules for these GPUs are frequently adjusted, reflecting their ongoing utility. A six-year-old A100, for example, retains a market value of about 25% of its original cost, indicating that older generations remain relevant and profitable. Fungibility is the third pillar, allowing NVIDIA's AI factories to support a diverse range of workloads. The ability to run various AI models, including proprietary and open-source, across different phases of development and deployment enhances the utility of these systems. This versatility is evident in numerous applications, from scientific computing to data processing, ensuring that the infrastructure remains valuable even as the nature of the work evolves. For example, companies like Lilly and Revolut leverage NVIDIA's technology for diverse tasks, from genomics modeling to processing vast amounts of transaction data. The implications of NVIDIA's advancements extend beyond immediate operational efficiencies. By creating a robust ecosystem that supports a wide range of applications, NVIDIA positions itself as a leader in the AI infrastructure market. Competitors may struggle to match the combination of productivity, durability, and fungibility that NVIDIA offers. As demand for AI capabilities continues to grow, operators will increasingly seek solutions that can adapt to changing workloads and maximize ROI over time. Looking ahead, NVIDIA's approach signals a shift in how companies will invest in AI infrastructure. The emphasis on general-purpose computing capabilities, coupled with ongoing software optimizations, suggests that operators will prioritize flexibility and long-term value over short-term gains. As the market evolves, those who can harness the full potential of NVIDIA's AI factories will likely emerge as leaders in the competitive landscape, driving innovation and efficiency across various sectors. The strategic focus on maximizing returns through versatile, durable, and productive AI systems will shape the future of AI deployment, compelling other players in the market to rethink their infrastructure strategies. ## Entities - **Companies**: NVIDIA, CoreWeave, Barkr, Silicon Data, Ornn Data, Lilly, Pinterest, Revolut, Runway, Texas A&M University, Cosm, Dassault Systèmes, Lucid Motors, Unilever - **Products**: NVIDIA A100 GPU, NVIDIA H100, NVIDIA Vera Rubin NVL72, NVIDIA GB300 NVL72, NVIDIA cuDF, NVIDIA Blackwell, NVIDIA Hopper, CUDA-X libraries - **Technologies**: AI factories, CUDA, agentic AI, deep learning, machine learning, generative AI - **People**: Jensen Huang ## Key Concepts Return on Investment, Earning Capacity, Useful Life, Demand, Productivity, Durability, Fungibility, General-purpose Accelerated Computing ## Definitions - **AI factories**: Facilities designed to optimize the production and deployment of AI models and workloads. - **CUDA**: A parallel computing platform and application programming interface model created by NVIDIA for general computing on GPUs. - **Earning Capacity**: The potential revenue a factory can generate based on its production capabilities. - **Durability**: The length of time that hardware remains economically viable and capable of generating revenue. - **Fungibility**: The ability of a system to run various types of workloads, enhancing its utility and revenue potential. ## Use Cases - Building and running protein models at Lilly - Post-training and deploying vision language models at Pinterest - Processing transaction data at Revolut - Training world models at Runway - Running molecular simulations at Texas A&M University - Providing flexible compute resources at Cosm ## Frequently Asked Questions **What is the cost of building an AI factory?** Each megawatt AI factory costs roughly $60 million. This investment is made with the expectation of a clear return on investment. **How does NVIDIA maximize the earning capacity of its AI factories?** NVIDIA maximizes earning capacity by delivering high throughput per megawatt and minimizing the cost per token, ensuring that factories can produce more revenue efficiently. **What is the useful life of NVIDIA's AI hardware?** NVIDIA's AI hardware, such as the A100 GPU, can remain economically viable for five to six years, with some systems continuing to earn revenue even longer. **What types of workloads can NVIDIA AI factories handle?** NVIDIA AI factories can run a wide range of workloads, including various AI models, data processing, scientific computing, and more, making them versatile and adaptable. **How does CUDA contribute to the longevity of NVIDIA hardware?** CUDA allows NVIDIA GPUs to run across generations, ensuring that existing hardware remains useful and capable of handling new workloads as they emerge. ## Links - [Read on Welcome.AI](https://welcome.ai/content/nvidia-ai-factories-enhance-revenue-with-high-throughput-and-durability) - [Original source](https://blogs.nvidia.com/blog/productive-durable-fungible-ai-factories/) --- Source: Welcome.AI | https://welcome.ai/content/nvidia-ai-factories-enhance-revenue-with-high-throughput-and-durability