Originally, there were quite a few positive messages over the weekend, but last night’s news directly caused me to reduce my fully invested SK Hynix position by half! 🧐

According to the latest news on the 26th, Nvidia is cutting its Rubin configuration significantly because memory is too expensive to sustain, reducing CPU memory by half, with no impact on the GPU’s HBM.

In terms of specific parameters, the SOCAMM memory module capacity in the rack has been reduced from 192GB directly to 96GB, and the total memory for the Vera CPU has been cut from 55TB to 28TB; the CPU rack portion has also switched to 96GB SOCAMM.

Previously, Nvidia relied on long-term storage agreements signed in advance, having been one of the companies most capable of enduring this price surge, but Apple was the first to succumb, and now even Nvidia is making cuts, demonstrating just how significant the pressure in the storage market is.

It should be noted that the Vera Rubin NVL72 is currently one of the most powerful AI systems, just released in January this year. According to actual test data, when processing MoE large model loads, the Blackwell platform achieves a throughput of 80,000 tokens per MW of power consumption, while the Vera Rubin NVL72 reaches 800,000 tokens per MW, a complete tenfold increase.

I hope the impact isn’t too great, after all, there's still half of the position left! 🥺

This article is sponsored by @binancezh, "Binance Buy American Stocks: Global Assets, Zero Time Difference, One-Click Access!"

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。