How Does an AI Factory Earn Back Its $60 Million Per Megawatt?
An AI factory is a data center built by the megawatt to produce tokens instead of physical goods. Each megawatt of capacity costs roughly $60 million to build, and operators commit that capital only when they can see a clear return.
The Three Numbers That Decide If an AI Factory Pays Off
Power is the binding constraint on an AI factory. This makes tokens per second per megawatt the number that governs earning capacity. More tokens inside a fixed power envelope means more revenue. Lower cost per token means more margin on it.
SemiAnalysis AgentX data shows NVIDIA Vera Rubin NVL72 systems deliver over 30x higher throughput per megawatt than NVIDIA GB300 NVL72. The same systems achieve up to 45x lower cost per million tokens on the DeepSeek V4 Pro model. Gains of that size come from extreme codesign across the full stack. Models and workloads are optimized together with software. Compute, networking, and memory are tuned as one unit.
Cheaper tokens do not shrink demand for compute. They expand it. Cheaper tokens make more use cases economical. Those use cases consume more tokens than the efficiency saved. Demand grows faster than efficiency improves.
Why the Newest Chip Doesn't Make the Old One Worthless
The NVIDIA A100 GPU shipped in 2020 and is still in commercial service six years later. CoreWeave recently extended bookings for units first introduced in 2020 through 2029. Every major operator has extended the depreciation schedule on its servers. That schedule is a guess about when hardware stops earning. The guess keeps moving out.
A September 2026 Sprout analysis, "The Productive Life of a Data Center GPU," tracks how that schedule has shifted across every major operator. Accounting life is a conservative proxy for physical life
The sources this article was written from
- NVIDIA Blog
https://blogs.nvidia.com/blog/productive-durable-fungible-ai-factories/