A single “memory compression algorithm” wiped over $10 billion in market value from two chip giants in one day. Overreaction, or the beginning of a rewrite for the AI memory demand story?
On Monday, Micron Technology (MU) closed at $321.80, down a sharp 9.88%. Trading volume surged to 72.4 million shares – nearly double its three‑month average of 36.3 million. Peer Sandisk (SNDK) fell 7.04% to $572.50, while Western Digital (WDC) tumbled 8.60%.
The only trigger? Google’s newly unveiled TurboQuant algorithm, released last week.
In a blog post, Google scientists described TurboQuant as a set of “advanced, theoretically grounded quantization algorithms” that enable massive compression for large language models and vector search engines. The most stunning claim:
At least 6x reduction in memory usage, up to 8x speedup, with zero accuracy loss.
Put another way – it could cut the number of memory chips needed by roughly 83%.
Cloudflare co‑founder and CEO Matthew Prince called it on X: “This is Google’s DeepSeek moment.” He added that there is “so much more room to optimize AI inference for speed, memory usage, power consumption, and multi‑tenant utilization.”
Early analysis suggests the impact is not uniform. NAND flash will bear the brunt, while DRAM and HBM (high‑bandwidth memory) are largely unaffected.
So in theory, Sandisk is at greater risk. But the market voted with its feet – both stocks were sold off together.
🐻 The Bear Case: Shrinking Demand → Lower Prices
Micron’s CFO Mark Murphy said on the Q2 earnings call:
If TurboQuant reduces NAND demand as promised, the supply‑demand balance will reverse. Prices would come under pressure, directly hurting Micron’s and Sandisk’s top lines.
🐂 The Bull Case: Cheaper Memory → More AI → Higher Total Demand
Some commentators argue that lower memory costs will spur AI adoption. As inference becomes cheaper, more businesses will deploy AI models, ultimately increasing total memory usage and demand.
This is reminiscent of the Jevons paradox – greater efficiency doesn’t necessarily reduce resource consumption; it can instead stimulate more use.
Over the past year, Micron had surged more than 250% on AI memory demand (especially HBM). Earlier this month, the company delivered blockbuster quarterly results. But since then, the stock had been trending downward as investors questioned:
TurboQuant is just the latest straw that broke the camel’s back.
In the short term, the real impact of TurboQuant remains unclear:
A few takeaways:
TurboQuant sounds scary, but it looks more like a catalyst to accelerate AI adoption than a death knell for memory chips. For Micron and Sandisk, short‑term pain is likely, but the long‑term outcome hinges on demand elasticity – and that is precisely what makes this battle so interesting.
