AI Data Center Construction Costs Reach Up to $39.5 Billion per Gigawatt

Building a 1-gigawatt artificial intelligence data center is estimated to require between $34.6 billion and $39.5 billion in capital spending, with costs varying by accelerator architecture.

Nvidia’s Vera Rubin architecture has the highest projected construction cost among the systems examined. OpenAI’s internally developed Jalapeno application-specific integrated circuit architecture ranks lower.

The projected cost of a single Nvidia Rubin NVL72 rack was reduced to $7.52 million from $9.1 million, a decline of about 17%. The revision reflects updated expectations for high-bandwidth memory prices and NAND storage capacity.

The estimates highlight the scale of AI infrastructure investment, which has also been examined in prior projections for AI infrastructure spending. For data center operators, the main economic burden is capital spending and the depreciation it generates rather than electricity costs.