Share.

    11 Kommentare

    1. WoodenHour6772 on

      As a regular consumer I’m pleased to see them get dinged a bit with this news, however one of the comments on that article points out it’s probably not as bad of an outcome as we’d like for these corporations:

      (E: idk what’s up with Reddit’s formatting, sorry for the weirdness)

      >Rebuttal: The Efficiency Paradox and the Memory Supercycle

      >The market’s negative reaction to Google’s TurboQuant algorithm is a classic case of short-term noise overriding long-term structural demand. The thesis that efficiency equals less demand is a fundamental misinterpretation of how the AI ecosystem scales.

      >1. The Jevons Paradox: Efficiency Explodes Demand
      History proves that when you make a resource (like memory) 6x more efficient, consumption does not drop—it skyrockets. By lowering the hardware barrier, Google’s breakthrough doesn t destroy the market; it democratizes it. High-performance AI can now move from massive data centers into billions of smartphones, laptops, and Edge devices. This creates a massive new Total Addressable Market (TAM) for Micron and SanDisk that previously didn t exist.

      >2. Physical Infrastructure vs. Software Recipes
      An algorithm is a recipe, but the chip is the essential physical infrastructure. Even as software optimizes storage, the OpenAI Stargate project—which requires trillions of dollars in investment—demonstrates that the sheer scale of future AI needs will always outpace software efficiency. As LYNX Equity Strategies noted, advanced compression is unlikely to hurt demand because we are currently limited by physical supply constraints, not software waste.

      >3. Sold Out is the Opposite of Bag Holding
      Labeling investors in Micron or SanDisk as bag holders due to a 3-6% dip is a financial absurdity.
      ​Micron has already confirmed that its high-bandwidth memory (HBM) is 100% sold out through 2026 under non-cancellable contracts.
      ​SanDisk is seeing record revenue growth driven by the shift to Agentic AI.
      You cannot be a bag holder of a critical commodity that the world s largest companies are standing in line to buy for the next two years.

      >Conclusion: A Volume Multiplier, Not a Threat
      Google’s breakthrough is a volume multiplier. It ensures AI is no longer a niche tool but a global utility. For Micron and SanDisk, this is the catalyst for the next wave of massive hardware deployment across the entire global economy. This dip is a healthy breather in a historic supercycle, not a change in trajectory.“

    2. beachfrontprod on

      People will just want to run bigger models faster no? It seems weird that the „market“ thinks that companies and users won’t still want to push it further once headroom appears.

    3. stuffitystuff on

      It doesn’t compress models, just the key-value store. It’s cool but no one will be able to use this to run a leaked Claude Opus or ChatGPT 5 on their laptops.

    4. The funny thing is I expect a lot of compression to be a major goal of AI. Much is bandwidth limited, so it’s very apparent that compression will become a major component of the AI process stack eventually.

    5. Not sure why they dropped. It’s nailed-on that no matter how much the data compession improves, storage will still be used and filled.

    6. I’m looking at my chrome browser and calling bullshit on Google being the ones with the memory use breakthrough.

    7. Ignorant knee jerk reaction by the market. AI context length is still dependent on high bandwidth memory. Google’s tech will just increase the thirst for longer context lengths.

    Leave A Reply