There were quite a few positive news updates over the weekend, but tonight’s news made me cut half of my SK Hynix position, which I had fully loaded! Latest update on the 26th: NVIDIA couldn’t handle the skyrocketing memory prices and drastically slashed Rubin configurations—CPU memory was halved, though GPU HBM remains unaffected. Looking at the specific parameters, the SOCAMM memory module capacity in the rack dropped from 192GB to 96GB, the total memory paired with Vera CPUs was cut from 55TB to 28TB, and the CPU rack section also switched to 96GB SOCAMM. Previously, NVIDIA was one of the companies best able to withstand this wave of price hikes, thanks to long-term storage agreements signed in advance. Apple was the first to cave, and now even NVIDIA has to make cuts, which shows just how much pressure the storage market is under. It’s worth noting that the Vera Rubin NVL72 is currently one of the most powerful AI systems out there, just launched this January. According to test data, when handling MoE large model workloads, the Blackwell platform achieves a throughput of 80,000 tokens/s per megawatt of power consumption, while the Vera Rubin NVL72 hits 800,000 tokens/s—a tenfold increase. Hopefully, the impact won’t be too severe, especially since I still have half my position! This post is sponsored by @binancezh: "Buy U.S. stocks on Binance—global assets, zero time lag, instant access!