As Google introduces more efficient AI compression technology, the 'shortage logic' in the memory industry has been challenged, dragging the sector and, in turn, the Nasdaq into a correction zone. However, analysts note that AI expenditure could exceed $1 trillion by 2030, with capital flows still signaling bullish sentiment.
In$Alphabet-A (GOOGL.US)$ Following the release of a technology considered 'potentially transformative for demand structure,' the global memory chip sector has experienced significant volatility recently. However, multiple institutions continue to emphasize that the long-term demand driven by artificial intelligence (AI) has not fundamentally changed.
Since March, the leading memory chip company $Micron Technology (MU.US)$ has seen its share price decline continuously, with a single-day drop of nearly 10% on Monday, and a cumulative retracement of approximately 30% from its mid-March peak. Despite the company's previously disclosed quarterly results showing revenue nearly doubling year-on-year, market sentiment has clearly weakened.
The direct factor triggering this round of sell-offs was Google's release of the AI compression algorithm TurboQuant. This technology compresses key-value caches (KV cache) during large model inference without compromising model accuracy, reducing memory usage by about six times and improving computational efficiency.
Market concerns lie in the possibility that improvements in AI computing efficiency could reduce demand for high-bandwidth memory (HBM) and DRAM, thereby undermining the core logic that had previously supported the rise of the memory sector. Following the announcement, $Micron Technology (MU.US)$ 、 $SanDisk (SNDK.US)$ 、 $Western Digital (WDC.US)$ and $Seagate Technology (STX.US)$ multiple storage-related stocks fell broadly, with related companies in South Korea and Japan also experiencing pressure.
Affected by this, the semiconductor sector as a whole weakened and further transmitted the impact to tech stocks. The Nasdaq Composite Index has recently entered a technical correction zone, indicating structural differentiation within AI-related trading.
However, several institutions believe this sell-off is more sentiment-driven rather than indicative of a fundamental turning point. Bank of America noted that similar compression and quantization technologies are not new and $NVIDIA (NVDA.US)$ relevant optimization solutions were introduced over the past year. The bank emphasized that the core indicator for assessing AI industry chain demand remains capital expenditure, not a singular efficiency improvement tool.
From the demand side, investment in AI infrastructure remains in an expansionary cycle. Institutions predict that global AI-related expenditures could exceed $1 trillion by 2030, with data centers' demand for memory continuing to rise. Meanwhile, supply in the industry remains tight, and some analysts expect the upward trend in DRAM prices to continue until 2027, with AI data centers becoming the primary source of memory demand.
In the longer term, efficiency improvements may even stimulate demand growth inversely. Some analysts cited the 'Jevons Paradox,' pointing out that cost reductions will expand AI application scenarios, thereby increasing overall computing and storage demand.
Currently, the global memory industry remains in a state of tight supply-demand balance. Since 2024, AI infrastructure construction has driven significant price increases in DRAM and NAND, with some niche products experiencing price surges of several multiples. Supply constraints are expected to persist until around 2027.
Against this backdrop, Bank of America maintains a positive rating for the memory sector, listing it as one of the key sub-sectors, second only to AI computing, semiconductor equipment, and networking. The bank believes that Micron Technology's current valuation is at the lower end of its historical range. Although short-term profit margins face uncertainty, there remains room for medium- to long-term recovery.
Editor/Doris