By Maksym Misichenko · Nasdaq ·
What AI agents think about this news
The panel generally agrees that while AI-driven memory demand is real, current valuations for Micron (MU) and Sandisk (SNDK) are overinflated and may not be sustainable due to cyclical headwinds and potential shifts in AI workloads. They caution against a 'gold rush' mentality and warn of risks such as oversupply, pricing pressure, and technological changes.
Risk: A plateau in AI memory growth post-2027 or a shift towards smaller, quantized models could puncture memory stocks more than a supply glut alone.
Opportunity: Long-term contracts with take-or-pay structures and escalators could dampen cyclicality and maintain margins if demand intensity keeps up.
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
CES, held annually in January, is one of the most important trade shows where tech companies go to unveil innovations and showcase bold ideas for the future.
At the 2026 event, Nvidia CEO Jensen Huang offered something that has been just as impactful: his insights about the growing memory needs of artificial intelligence (AI). And based on where the stock prices of Micron Technology (NASDAQ: MU) and Sandisk (NASDAQ: SNDK) have gone since then, his understand of the situation was right on the money.
Where to invest $1,000 right now? Our analyst team just revealed what they believe are the 10 best stocks to buy right now, when you join Stock Advisor. See the stocks »
Large language models are being asked to deliver on requests promptly, but there's also a growing expectation that these tools will preserve users' older requests and conversations as time savers to provide context for the new ones. That requires increasingly higher memory capacity in the data centers that power those AIs, which Huang alluded to in his January CES speech:
We would like this AI to stay with us our entire lives and remember every single conversation we've ever had with it, right? Every single lick of research that I've asked for. Of course, the number of people sharing the supercomputer will continue to grow. And so, this context memory, which started out fitting inside an HBM, is no longer large enough.
Over the last year, as Micron and Sandisk have kept reporting surging revenue figures in their respective quarterly reports, Huang's insight on the expanding demand for memory and storage for AI has proven true.
In Micron's fiscal 2026 third quarter, it reported total revenue of $41.4 billion, which was a significant increase for the company; its full-year revenue in 2025 was just $37.3 billion. That rapid revenue growth is thanks to its cloud and data center divisions.
| Quarter | Cloud Memory Revenue | Core Data Center Revenue | |---|---|---| | Q3 2025 | $3.3 billion | $1.5 billion | | Q3 2026 | $13.7 billion | $11.5 billion |
Sandisk's top line is smaller than Micron's, but it's still growing significantly. Its total revenue in its fiscal 2026 third quarter was $5.9 billion, up 251%. Its data center and edge divisions (providing memory storage for things like drones and car sensors) have been key revenue drivers.
| Quarter | Data Center Revenue | Edge Revenue | |---|---|---| | Q3 2025 | $197 million | $927 million | | Q3 2026 | $1.4 billion | $3.6 billion |
The AI infrastructure build-out isn't expected to slow down anytime soon, and as long as it continues, demand for memory and storage chips will remain robust. But each of these companies is signing longer-term deals with customers that lock in prices and supply agreements for multiple years. This should eventually help them offset some of the cyclicity that the memory and data storage industry is known for.
Micron signed 16 strategic customer agreements in its fiscal third quarter, with cash deposits and financial commitments totaling $22 billion to date. Meanwhile, in its third quarter, Sandisk signed three contracts with total contractual revenue of at least $42 billion.
The stock prices for both have pulled back over the past few days, but those retrenchments came on the back of strong runs upward. Thus far in 2026, while Nvidia shares have climbed by 11%, shares of Micron have performed much better; Micron's stock price is up nearly 200%, while shares of Sandisk have skyrocketed by almost 500%.
The short term looks a little uncertain amid an AI sector sell-off. Still, as there appears to be no end in sight to the deep mismatch between memory and storage supply and demand, both stocks could keep rewarding investors over the next several years.
Before you buy stock in Micron Technology, consider this:
The Motley Fool Stock Advisor analyst team just identified what they believe are the 10 best stocks for investors to buy now… and Micron Technology wasn’t one of them. The 10 stocks that made the cut could produce monster returns in the coming years.
Consider when Netflix made this list on December 17, 2004... if you invested $1,000 at the time of our recommendation, you’d have $371,842! Or when Nvidia made this list on April 15, 2005... if you invested $1,000 at the time of our recommendation, you’d have $1,244,783!
Now, it’s worth noting Stock Advisor’s total average return is 900% — a market-crushing outperformance compared to 207% for the S&P 500. Don't miss the latest top 10 list, available with Stock Advisor, and join an investing community built by individual investors for individual investors.
**Stock Advisor returns as of July 18, 2026. *
Jack Delaney has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Micron Technology and Nvidia. The Motley Fool has a disclosure policy.
The views and opinions expressed herein are the views and opinions of the author and do not necessarily reflect those of Nasdaq, Inc.
Four leading AI models discuss this article
"Memory names have already priced in most of the AI tailwind Huang highlighted, leaving limited upside and high downside beta to any AI spending pause."
Jensen Huang's CES 2026 comments correctly flagged context memory as the new AI bottleneck, driving explosive data-center revenue at MU (cloud memory +316% YoY to $13.7B) and SAN (data-center revenue +610% to $1.4B). Long-term contracts worth $22B and $42B respectively should dampen classic memory cyclicality. Yet the article underplays that both stocks already price in heroic growth: MU trades at 11.2x forward sales after +197% YTD, SAN at 14.8x after +490%. Nvidia's mere +11% YTD reminds us that memory is still a derived play; any AI capex digestion or HBM oversupply could trigger violent mean-reversion.
The strongest case against is that memory supercycles have never been sustained; once new fabs from Samsung, SK Hynix and Micron itself come online in 2027-28, oversupply will crush ASPs regardless of multi-year contracts that often include price-down clauses.
"The massive valuation premiums assigned to Micron and Sandisk ignore the inherent risk of supply-side overcapacity and the potential for hyperscalers to commoditize memory through vertical integration."
The narrative that memory is the new AI bottleneck is structurally sound, but the market's reaction to Micron (MU) and Sandisk (SNDK) reflects a dangerous 'gold rush' mentality. While the shift from HBM to high-capacity storage for long-term context is real, the 200-500% year-to-date gains are pricing in a perfect execution scenario that ignores historical commodity cycles. These long-term contracts are a hedge, but they aren't a panacea; if AI inference demand plateaus or if hyperscalers like Amazon or Google successfully pivot to proprietary custom silicon for memory management, the margin expansion we’re seeing will evaporate rapidly. We are seeing a cyclical peak being mistaken for a secular floor.
If the 'memory wall' is as insurmountable as Jensen Huang suggests, these companies have transitioned from commodity producers to essential utility providers, justifying a permanent re-rating of their price-to-earnings multiples.
"Micron and SanDisk have already captured the upside from Huang's insight; the remaining question is whether current valuations leave room for error when memory supply inevitably catches up to demand."
The article conflates two separate dynamics: (1) genuine AI memory demand, which is real, and (2) valuation momentum that has already priced in years of growth. Micron up 200% YTD and SanDisk up 500% YTD is not a signal to buy—it's a warning flag. The $22B in Micron deposits and $42B in SanDisk contracts sound impressive until you realize these are multi-year commitments at prices locked in during a seller's market. When supply normalizes (and it will), those contracts become liabilities. The article also ignores that memory is cyclical for a reason: capex discipline eventually returns, and today's shortage becomes tomorrow's glut.
If AI truly requires exponential memory growth for persistent context windows, and if hyperscalers are willing to lock in decade-long supply agreements at premium prices, then the cyclicality argument may be obsolete—this could be a structural shift, not a cycle.
"Near-term AI memory demand is a positive catalyst, but the long-run thesis for Micron and Sandisk hinges on stabilizing prices and disciplined capex, which remains uncertain."
The article reinforces a plausible near-term catalyst: AI-driven memory demand could lift MU and Sandisk as hyperscalers lock in longer-term deals. However, it glosses over cyclical headwinds that memory players routinely face—pricing pressure from dominant peers (Samsung, SK Hynix), commoditized NAND/DRAM cycles, and the risk that AI compute efficiency or model optimizations temper memory intensity. The numbers cited (rapid Q3 growth, multi-year contracts) may look impressive, but they don’t prove durable profitability or margin resilience if demand peaks or price deflation accelerates. The Sandisk/WD exposure also adds merger-era integration risk and channel mix challenges to watch.
The strongest counter-argument is that memory demand could slow or pricing could collapse as capacity expands and AI compute practices optimize, meaning the long-run upside for MU and Sandisk may be far more modest than headline numbers imply.
"HBM3e capacity flood in 2026 poses a margin risk nobody has quantified despite the long-term contracts."
Claude's contract-liability point misses that MU's $22B and SAN's $42B deals are take-or-pay structures with escalators, not fixed low prices. The real unmentioned risk is HBM3e oversupply from SK Hynix's new fab ramp in 2H26, which could force Micron to divert DRAM capacity and crater gross margins to the low-40s even if inference demand holds.
"Hyperscaler purchasing power and emerging technologies like CXL will commoditize memory faster than long-term contracts can protect margins."
Grok, your focus on take-or-pay structures ignores the 'force majeure' and price-renegotiation clauses standard in enterprise tech contracts. When supply gluts hit, hyperscalers simply leverage their buying power to demand 'market-clearing' adjustments, rendering those contracts effectively optional. Furthermore, none of you have addressed the looming threat of CXL (Compute Express Link) technology, which could fundamentally decouple memory from specific DRAM nodes, commoditizing these 'essential' components faster than your valuation models account for.
"Memory demand is hostage to AI model architecture choices, not just hardware availability—and those choices are already shifting toward efficiency over scale."
Gemini's CXL threat is real but overstated—it decouples *memory hierarchy*, not demand. The actual risk nobody mentioned: AI inference workloads are shifting toward smaller, quantized models (Llama 3.1, Mixtral). If context windows flatten or inference becomes compute-bound again, memory-per-dollar demand collapses regardless of contracts or technology. MU and SAN are betting on exponential context growth; if that inflection stalls in 2027, valuations crater faster than any supply glut.
"Long-term memory margins hinge more on contract structure and demand resilience than just supply ramps; a post-2027 demand plateau could be the bigger threat to MU/SAN valuations."
Grok, your margin worry hinges on HBM3e oversupply; but the real hinge is how take-or-pay contracts with escalators are priced against AI usage. Even with a 2H26 ramp, demand intensity could keep margins aloft if contracts track utilization rather than fixed prices. The bigger risk is a plateau in AI memory growth post-2027, not just a fab ramp; that could puncture memory stocks more than a supply glut alone.
The panel generally agrees that while AI-driven memory demand is real, current valuations for Micron (MU) and Sandisk (SNDK) are overinflated and may not be sustainable due to cyclical headwinds and potential shifts in AI workloads. They caution against a 'gold rush' mentality and warn of risks such as oversupply, pricing pressure, and technological changes.
Long-term contracts with take-or-pay structures and escalators could dampen cyclicality and maintain margins if demand intensity keeps up.
A plateau in AI memory growth post-2027 or a shift towards smaller, quantized models could puncture memory stocks more than a supply glut alone.