Reminiscence shares received hammered this week after Google dropped a analysis paper that has traders questioning all the thesis for the AI-driven reminiscence bull run. Alphabet’s (GOOGL) Google Analysis group revealed particulars on Tuesday of a brand new compression algorithm known as TurboQuant, and the fallout was swift. Sandisk Company (SNDK) plunged as a lot as 11%, Micron Know-how (MU) dropped 7% on Thursday because the promoting accelerated, and Western Digital and Seagate every fell over 7%. The harm prolonged abroad, with Samsung and SK Hynix each sliding greater than 5% in Seoul. As of this morning, the group is rebounding.
Each Micron and SanDisk at present carry a Zacks Rank #1 (Sturdy Purchase). The availability constrained dynamic and distinctive demand for DRAM has pushed these shares to unbelievable performances within the final six months, with SNDK up almost 5x and MU greater than doubling over that interval.
So is Alphabet constructing a expertise that destroys the bull case for this area of interest AI infrastructure increase? Or is that this a case of headline-driven profit-taking in an overextended sector?
Picture Supply: Zacks Funding Analysis
What Alphabet’s TurboQuant Truly Does
At its core, TurboQuant addresses one of the vital costly bottlenecks in operating giant language fashions: the key-value (KV) cache. That is the high-speed knowledge retailer that retains context so a mannequin would not must recompute every little thing with every new token it generates. As fashions course of longer inputs, the KV cache balloons, consuming GPU reminiscence that would in any other case serve extra customers or run bigger fashions.
Google’s algorithm compresses the KV cache considerably, decreasing its reminiscence footprint by almost 6x with out sacrificing accuracy or requiring mannequin retraining. Examined throughout 5 customary AI fashions, TurboQuant achieved good scores on retrieval duties. Testing confirmed that it delivered as much as an 8x acceleration in computing consideration on Nvidia H100 GPUs.
Past LLMs, Alphabet famous that TurboQuant additionally improves vector search, the expertise underpinning every little thing from Google Search to YouTube suggestions to advert focusing on.
Reminiscence Shares Micron and Sandisk Promote Off Sharply
The sell-off in Micron, Sandisk, and their reminiscence friends echoed the DeepSeek-driven panic from early 2025, when a Chinese language AI lab demonstrated that aggressive fashions might be educated with far much less compute than assumed. That episode triggered a one-day bloodbath in semiconductor names earlier than the market in the end concluded that effectivity beneficial properties speed up adoption, and adoption drives extra {hardware} demand, not much less.
The identical logic might apply right here. TurboQuant compresses knowledge throughout inference, not coaching. That is an necessary distinction. AI mannequin coaching, which is the place probably the most reminiscence consumption happens, is basically unaffected by this algorithm. Excessive Bandwidth Reminiscence (HBM), the product class driving the strongest income progress at Micron and its friends, stays important for coaching workloads.
There’s additionally the Jevons paradox to think about: by making AI inference cheaper and extra memory-efficient, TurboQuant may speed up the deployment of LLMs into edge gadgets, smartphones, and IoT functions, all classes that will create totally new reminiscence demand that does not exist at present.
SNDK and MU Inventory’s Nonetheless Boast Tailwinds
Micron’s earnings final week have been fairly distinctive. The corporate delivered February-quarter earnings that beat steering by 45%, and its Might-quarter earnings steering successfully doubled consensus estimates. Income almost tripled year-over-year, and internet revenue surged virtually 10x. MU at present trades at roughly 6x ahead earnings, the most cost effective ahead P/E in all the S&P 500 after its latest 23% post-earnings slide from report highs.
SanDisk’s fundamentals inform an identical story. SNDK reported Q2 fiscal 2026 EPS of $6.20, beating forecasts by almost 78%, with income up 61% year-over-year. Gross margins expanded dramatically to 51.1% from 29.9% within the prior quarter, and administration guided Q3 income to $4.4–$4.8 billion with gross margins of 65–67%. The corporate is actively signing long-term provide agreements and projecting 75–100 exabytes of AI-related storage demand by 2027.
Whereas investor focus stays fixated on Alphabet’s analysis breakthroughs, the extra related danger to those shares isn’t technological obsolescence pushed by analysis, however the inherently cyclical nature of the reminiscence and storage trade itself. Traditionally, these companies have been outlined by intervals of undersupply adopted by aggressive capability enlargement, which in the end results in pricing strain and margin compression. In that context, the eventual finish of this cycle is much extra prone to be pushed by supply-demand dynamics normalizing than by any single leap ahead in AI capabilities.
Within the chart under, we are able to see that Micron inventory has traded all the way down to a transparent stage of assist following the earnings, and Alphabet information pushed selloff. For now, it seems to be holding at assist. Traders can regulate this stage to find out the place the subsequent transfer could also be.
Picture Supply: TradingView
The AI Buildout Exhibits Few Indicators of Slowing
Alphabet’s TurboQuant is a official technical achievement. It should possible cut back the price of AI inference over time, and GOOGL stands to learn instantly by way of decrease working prices throughout its cloud and search infrastructure. However the leap from “6x compression of the KV cache” to “the AI reminiscence increase is over” could also be a story stretched too far.
The reminiscence cycle is pushed by supply-demand dynamics that function on multi-year capex timelines, not by particular person analysis papers. Samsung, SK Hynix, and Micron have all constrained provide in latest quarters exactly as a result of the trade realized its lesson from the brutal 2022–2023 downturn. Demand from hyperscale knowledge middle buildouts, the place Meta alone just lately dedicated $27 billion in a single cope with Nebius, exhibits no indicators of abating.
For traders in Micron and SanDisk, this week’s pullback seems to be extra like a present than a warning. Each shares are Zacks Rank #1 names with accelerating earnings, strengthening margins, and valuations which have moderated within the final couple of weeks.
Simply Launched: Zacks Prime 10 Shares for 2026
Hurry – you’ll be able to nonetheless get in early on our 10 high tickers for 2026. Handpicked by Zacks Director of Analysis Sheraz Mian, this portfolio has been stunningly and constantly profitable.
From inception in 2012 by way of November, 2025, the Zacks Prime 10 Shares gained +2,530.8%, greater than QUADRUPLING the S&P 500’s +570.3%.
Sheraz has combed by way of 4,400 firms coated by the Zacks Rank and handpicked the very best 10 to purchase and maintain in 2026. You’ll be able to nonetheless be among the many first to see these just-released shares with huge potential.
Micron Know-how, Inc. (MU) : Free Inventory Evaluation Report
Sandisk Company (SNDK) : Free Inventory Evaluation Report
Alphabet Inc. (GOOGL) : Free Inventory Evaluation Report
This text initially revealed on Zacks Funding Analysis (zacks.com).
The views and opinions expressed herein are the views and opinions of the writer and don’t essentially mirror these of Nasdaq, Inc.
