The full-scale production rollout of Nvidia's next-generation AI platform, Vera Rubin, is putting pressure on the TLC NAND supply chain, pushing the spot price of 512Gb TLC NAND, which fell in June, back above $21.
As AI models grow larger, KV Cache* management has emerged as a key challenge, and Nvidia has introduced a new intermediate storage layer called CMX (Context Memory eXtension) to the Rubin platform to address this.
* Key-Value Cache is a memory space where AI models temporarily store calculated information for reuse.
CMX sits between HBM (high bandwidth memory) and existing backend network storage, operating as a high-speed TLC flash memory pool.
* 1 unit CMX server can accommodate 600TB of TLC flash memory
This is expected to expand the benefits for memory semiconductor suppliers (Samsung Electronics, SK Hynix, etc.).
- The upward trend in prices (price rebound and improved profitability) in the HBM market as well as the high-performance TLC NAND and eSSD markets is likely to accelerate.
Therefore, memory companies with strong capabilities in supplying CMX servers and high-density TLC chips for enterprise SSDs are expected to further strengthen their position.
However, while the hypothesis that "Nvidia is driving up TLC NAND prices" is quite convincing, there are also opinions that more verification is needed at this stage. Please take note!
▶ Original source: https://wccftech.com/nvidias-vera-rubin-production-ramp-up-is-now-squeezing-tlc-nand-supply-driving-512gb-spot-prices-to-21-after-the-june-slump/
