Key insights
- The author argues that misinterpretations of AI research papers (Deepseek, TurboQuant) trigger unwarranted sell-offs in memory stocks, causing retail investors to panic and exit the market. This erodes trust in investing and creates unhealthy volatility. While long-term investors see buying opportunities, the author suggests the long-term impact of these events is negative for market participation.
Recently Google teased TurboQuant (an algorithm that compresses memory) so the market thought that demand will crash for RAM so let’s just sell everything. So we all saw how tens of billions were gone from memory stocks.
This is now the second time in 1 year the exact same trade played out over an AI research paper nobody actually reads by the way. January 2025, Deepseek (the chinese AI model) dropped and we saw the same panic,the same thesis but stocks recovered as all. The same story will play out this time too. Both times the thesis was wrong and not in a complicated hard to see way. With Deepseek, cheaper inference expanded who could afford AI, demand went up not down. With TurboQuant, the algorithm only touches inference memory and has zero impact on training memory where the majority of HBM demand actually comes from. The paper has been sitting since 2025 and isn't even deployed at Google yet. This wasn't a hidden information by the way, it was right there in the research.
Yes, long term investors got a buying opportunity both times. But that's not the whole picture. Every time this happens retail investors who don't understand why their portfolio dropped 7% in two days panic sell at the bottom. They take real losses and they don't come back. And the next time something like this happens more people sit out entirely because they remember what happened last time.
Market pricing in fear over news nobody reads is not healthy volatility. It's just noise that slowly erodes trust in investing altogether and the people who pay for it are never the ones causing it.
Adding a link if you want the full breakdown of what this TurboQuant actually does. Here.