On October 10, Bernstein's research report revealed that the capital expenditure required to build a 1GW data center using different AI accelerator architectures ranges from approximately $34.6 billion to $39.5 billion. Among these, the construction cost of NVIDIA's Vera Rubin architecture is the highest, while OpenAI's self-developed ASIC architecture, Jalapeno, is relatively lower. Bernstein significantly revised its cost expectation for NVIDIA's Rubin NVL72 single rack from $9.1 million to $7.52 million, a reduction of about 17%, primarily reflecting adjustments in expectations for HBM prices and NAND storage capacity. The report also pointed out that the main economic burden of AI data centers is not electricity costs, but rather the substantial capital expenditures and the depreciation they generate.
All Comments