Chip memory stocks (Micron, Samsung, SK Hynix, etc) have significantly outperformed over the last twelve months due to undersupply concerns and the prioritization of AI data centres over consumer electronics demand. That said, last week suffered significant losses Micron, SK Hynix or Samsung have been underperforming technology peers:
Multiple factors are behind this move such as USD dynamics, geopolitical-related international trade disruptions and, most importantly, a new industry-related breakthrough that assuage the projected undersupply unbalanced: Google's TurboQuant.
TLDR: TurboQuant is a mathematically clever compression method that lets AI models run faster, cheaper, and with far less memory, without losing accuracy.
Key Takeaways:
- TurboQuant is a compression algorithm for AI models, specifically designed to reduce how much memory (RAM/VRAM) large models need while running.
- TurboQuant targets LLMs (like Gemini, GPT-style models) focusing on inference, not training
- TurboQuant achievements:
- ~6× less memory usage (confirmed by ML community tests)
- Up to ~8× faster processing (still contested by ML community tests).
- No loss in accuracy (this is the big deal)
- Impact:
- 1) Cheaper AI infrastructure: Less GPU memory needed, Lower cloud costs
2) Bigger context windows: AI can handle longer documents/conversations
3) Edge AI becomes possible: Phones / smaller devices can run stronger models
4) Industry ripple effects: Could reduce demand for high-end memory chips. Already impacted semiconductor stocks
More technical details in Google Research: TurboQuant: Redefining AI efficiency with extreme compression
| Research |
remove preview |
|
| TurboQuant: Redefining AI efficiency with extreme compression |
| Vectors are the fundamental way AI models understand and process information. Small vectors describe simple attributes, such as a point in a graph, while "high-dimensional" vectors capture complex information such as the features of an image, the meaning of a word, or the properties of a dataset. |
| View this on Research > |
|
|
------------------------------
Carlos Salas
Portfolio Manager & Freelance Investment Research Consultant
------------------------------