What Happened
Google unveiled TurboQuant, a technology that dramatically improves AI inference efficiency. It increases KV cache capacity by 6x and makes AI inference 8x faster with no accuracy loss. The core mechanism converts complex multidimensional vector embeddings into simplified directional commands with 1-bit error correction.
Markets reacted immediately. Micron (MU) fell 15% over five days, now 28% below its 52-week high of $471. DRAM spot prices also declined roughly 6% since Micron's recent earnings report.
Why This Combination Matters
Viewed through a single article, the story is simple: "Google makes AI use less memory, memory stocks fall." But combining four sources reveals a more complex picture.
Bear case (Citi): Cut price target from $510 to $425. DRAM weakness and AI efficiency trends pressure near-term demand.
Bull case (BoA, JPM): BoA analyst Vivek Arya argues "AI capital expenditures are driving memory usage, not efficiency measures." Micron's latest quarter backs this up: revenue $23.86B (nearly 3x year-over-year), EPS $12.07 (up from $1.41), gross margins above 74%. The numbers say AI memory demand is structural.
The Jevons Paradox debate: Motley Fool's Billy Duberstein draws parallels to the DeepSeek moment earlier this year. Then, too, fears emerged that "more efficient AI means less chip demand." In reality, companies reinvested savings into building more advanced models. Efficiency gains led to cost reduction, then more usage, then higher total demand. The same pattern could repeat.
Risk
Jevons Paradox doesn't always hold. If TurboQuant triggers a demand shift from HBM to DDR5/MRDIMM, it directly hits Micron's high-margin HBM business. And if DRAM spot price declines extend into Q2, earnings downgrades become inevitable.
Investment Implications
- Micron holders: high short-term volatility, but BoA consensus target of $494 (+46% from current) remains in play if Q2 DRAM pricing stabilizes
- TurboQuant's actual adoption speed is the key variable. There's a gap between tech announcement and commercial-scale deployment
- DeepSeek lesson: watch whether the "initial panic sell then rebound" pattern repeats