Productive, Durable, Fungible: How NVIDIA AI Factories Maximize Return on Investment
AI factories are built by the megawatt, even by the gigawatt. Each megawatt factory costs roughly $60 million, and AI factory operators will only commit capital on that scale with a clear view of the return on investment. Three key things shape AI factory returns: Earning capacity: What the factory could earn in a year […]
NVIDIA AI factories are constructed on a grand scale, often reaching gigawatt levels. A significant investment of $60 million per megawatt prompts operators to commit capital only after considering the potential return on investment. Three factors contribute to AI factory success: earning capacity, useful life, and demand. A factory's earning capacity is the amount it could generate in a year if it sold all its tokens.
Its useful life refers to the duration its hardware remains profitable. Demand signifies the market need for those tokens. No single factor can compensate for weaknesses in others.
Productivity is crucial, as it determines the factory's earning capacity. Lower cost per token not only boosts margin but also expands the market for the factory. SemiAnalysis AgentX data reveals NVIDIA's Vera Rubin NVL72 systems surpass NVIDIA's GB300 NVL72 by over 30 times in throughput per megawatt and provide up to 45 times lower cost per million tokens on the DeepSeek V4 Pro model.
This productivity is achieved through comprehensive codesign across the entire system, encompassing models, workloads, software, compute, networking, and memory, all optimized in unison.
Durability is essential because demand doesn't always align with technological advancements. The installed base of older hardware continues to generate revenue, as demonstrated by the NVIDIA A100 GPU, which, despite being released in 2020, remains in commercial service six years later. Major operators consistently extend the depreciation schedule of their servers, which indicates a growing belief that hardware maintains earning potential beyond the manufacturer's stated life.
Fungibility plays a pivotal role in AI factories, as NVIDIA's systems cater to a wide range of AI workloads, as well as numerous non-AI applications. This broad applicability deepens and broadens the demand they can serve. NVIDIA's platform demonstrates the power of productive, durable, and fungible design, maximizing AI factory returns through engineering codesign, continuous software optimization, and a standardized architecture accessible to any operator, deployable from a validated reference design.
Written by urgent.news from NVIDIA Blog's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
