[Bernstein breaks down the cost of AI infrastructure: up to US$39.5 billion per gigawatt of investment] On October 10, a Bernstein research report showed that using different AI accelerator architectures, the capital expenditure required to build a 1GW data center is approximately US$34.6 billion to US$39.5 billion. Among them, Nvidia's Vera Rubin architecture has the highest construction cost, while OpenAI's self-developed ASIC architecture Jalapeno is relatively low. Bernstein significantly lowered Nvidia's Rubin NVL72 single-cabinet cost forecast to $7.52 million from the previous $9.1 million, a drop of about 17%, mainly reflecting adjustments to HBM price and NAND storage capacity expectations. The research report also pointed out that the main economic burden of AI data centers is not the electricity bill, but the huge capital expenditure and the depreciation it generates.