According to CoinMeta, the Vera Rubin NVL72 of Nvidia achieved a throughput of up to 3.7 times in the latest V6.1 testing of MLPerf, setting a benchmark for AI hardware performance. However, with the acceleration of the hardware cycle, AI companies are facing increasing financial pressure. Nvidia indicates that the 288 GPU in four GB300 NVL72 racks have reached 99% expansion efficiency, and software improvements have also brought a 1.6-fold increase in performance. At the same time, the MI350 series of accelerators from AMD are also standing out in the competition. CoreWeave recently expanded convertible note financing from $3 billion to $3.7 billion, demonstrating the urgent need for funds in the AI infrastructure industry. Reuters reports that AI related borrowers are facing higher interest rate spreads, and investors are expressing concerns about the industry's future capital needs. Nvidia believes that more throughput can reduce the cost per token, and future hardware investments may be rewarded with better economics.