Micron vs SanDisk: Who's Winning the AI Memory Race?

On June 22, 2026, Micron (MU) and SanDisk (SNDK) hit all-time highs in the same session. Micron closed at $1,211.38, a fresh record, the same day it announced a strategic multi-year partnership with Anthropic covering HBM supply, co-design, and an investment in Anthropic's Series H round. SanDisk closed at $2,273.73, also a record, touching an intraday high of $2,354.90. By June 23, both were down over 13%, caught in a global sell-off that started in Seoul, South Korea and spread to every AI-adjacent stock on the planet. 

Here is the thing most people get wrong: these two companies are not racing each other. Micron owns one layer of the AI memory stack, SanDisk owns a completely different one. Understanding that difference is the key to understanding what happened, why both stocks have done what they have done, and how to think about them going forward.

Let's break down the business of each company, the reason both stocks have delivered some of the best returns in the S&P 500 this year (MU up 233% YTD, SNDK up 613% YTD as of the June 23 close), what triggered the synchronized crash, what the fundamentals and valuations actually look like after the dip, and what analysts think about each.

Micron vs SanDisk: How Their AI Memory Businesses Are Different

The easiest mistake is treating Micron and SanDisk as head-to-head competitors. They are not. Both benefit from the same AI-driven demand for memory, both are US-listed pure-play memory companies, and both get traded together by momentum investors. But they build fundamentally different products for fundamentally different roles inside an AI data center.

Here is the simplest way to think about it. Every smartphone you own has two types of storage: RAM and internal storage. RAM is fast, keeps active processes running, and gets wiped when you restart. Internal storage is slower relative to RAM but permanent, holding your photos, apps, and data. AI data centers work the same way at a different scale. Micron makes the AI equivalent of RAM: DRAM and its premium version, High-Bandwidth Memory (HBM), which is physically stacked inside Nvidia's GPUs to enable model training. SanDisk makes the AI equivalent of internal storage: NAND flash chips and enterprise solid-state drives (SSDs) that hold the model weights, training datasets, and inference caches that AI systems need to keep.

SanDisk was spun off from Western Digital in February 2025 at approximately $38.50 per share. It joined the Nasdaq 100 in April 2026. Before the spinoff, the NAND business was buried inside a hard-drive conglomerate. Once separated, investors could price it as a pure-play AI storage company, and the market re-rated it violently. From the spinoff price to the June 22 all-time high, SNDK returned approximately 6,000%, one of the most extraordinary single-stock stories in recent semiconductor history. Micron's story is different: it is a decades-old company entering the most profitable period in its history, having crossed $1 trillion in market cap this year for the first time.

Here is how the two businesses compare at a glance:

The gross margin figures here deserve a second look. Both companies are now printing software-like margins on semiconductor hardware, which reflects just how severe the supply-demand imbalance is for AI memory. In normal memory cycles, NAND gross margins run in the 30 to 40% range. SanDisk hitting 78.4% tells you something extraordinary is happening on the pricing side.

Why Micron and SanDisk Stocks Have Rallied So Much

Short version: AI is consuming memory faster than anyone can make it.

Every large language model running at scale requires memory at multiple layers. Training a frontier AI model needs enormous amounts of HBM stacked inside GPU accelerators. Running inference, the part where a model answers your query, needs fast enterprise SSD storage to hold model weights and serve results efficiently. Hyperscalers including Microsoft, Google, Meta, and Amazon collectively committed more than $350 billion in AI data center capex in 2026. All of that flows downstream into chips, and memory captures a significant share of that spend.

Supply is the other half. New semiconductor fab capacity takes 18 to 24 months minimum to come online. Goldman Sachs estimated the 2026 DRAM supply-demand gap at 4.9%, the worst shortage in 15 years. NAND supply growth is running at roughly 15 to 17% this year against demand growth of 20 to 22%. When demand outpaces supply in a high fixed-cost business, margins don't expand gradually. They spike.

The structural change that makes this cycle different from previous ones is the shift to multi-year supply agreements. Micron confirmed on its Q1 FY2026 earnings call that its entire 2026 HBM supply is sold out under contract at locked prices. SanDisk disclosed five multi-year supply agreements, which the company calls New Business Models (NBMs), covering $42 billion in minimum contracted revenue backed by $11 billion in enforceable financial guarantees. These commitments look less like commodity chip sales and more like the long-term infrastructure contracts that hyperscalers sign with data center developers.

SNDK's 613% YTD return partly reflects the pure-play re-rating effect. When NAND was inside Western Digital's conglomerate, it was valued like a hard-drive business. Once separated, it got priced as an AI storage infrastructure company. MU's return, extraordinary on its own terms, looks moderate by comparison because Micron was already a large-cap company before 2026. Different starting points, same AI tailwind.

Why Micron and SanDisk Stocks Fell on June 23, 2026: Blame Seoul, Not the Stocks

On June 23, South Korea's KOSPI index crashed nearly 10%, triggering a circuit breaker halt. Samsung Electronics and SK Hynix, the two largest Korean memory chipmakers and together responsible for roughly 70% of the KOSPI's 2026 gains, both fell more than 12%. The immediate triggers, as per reporting from Reuters, Bloomberg, and CNBC, included concerns about AI competitiveness following leadership departures at Google and regulatory scrutiny over leveraged semiconductor-linked financial products in Korean markets.

The contagion moved fast. US memory stocks followed Asian markets lower in premarket. The Nasdaq closed down 2.21%. Micron fell 13.18% from its June 22 close of $1,211.38 to $1,051.77. SanDisk fell 13.64% from $2,273.73 to $1,963.60.

There is a useful way to read this. Micron and SanDisk had both hit all-time highs on June 22, including the same day Micron announced the Anthropic partnership, which had pushed sentiment to peak optimism. A global macro shock gave momentum traders the permission they were already looking for to take profits. Wedbush analyst Dan Ives, quoted on CNBC, characterized the drop as 'sell the news, profit-taking after an extreme rally, not a full crash.' Several other analysts echoed the same view.

Critically, nothing changed about the businesses on June 23. No contracts were cancelled. No earnings estimates were revised downward. BofA's Vivek Arya chose that exact session to raise his Micron price target from $950 to $1,500, calling the dip a structural buying opportunity and arguing that DRAM and HBM are now mission-critical AI infrastructure, not commodity cycle plays. That kind of public call on a down day is not something analysts do casually.

Micron's Earnings Today: The Live Market Stress Test

Micron is reporting Q3 FY2026 earnings today, June 24, after the US market close. The results will be the single most watched data point in the AI semiconductor sector right now.

Street consensus heading into the print: revenue of approximately $34.5 to 35.6 billion and non-GAAP EPS near $19.72, against Micron's own guidance of $33.5 billion in revenue and $19.15 in non-GAAP EPS. Analysts expect a beat. Micron has beaten consensus EPS in each of its last four quarters, averaging a 21.7% positive surprise.

For context on how far the business has moved in a year: in Q3 FY2025, Micron reported $9.3 billion in revenue and $1.91 EPS. The year-over-year growth expected this quarter is roughly 271% on revenue and 932% on EPS. Analysts at TradingKey note that this growth is driven by both HBM ramp (Micron began high-volume production of HBM4 for Nvidia's Vera Rubin platform earlier this year) and sharp DRAM and NAND pricing increases.

What Wall Street is actually focused on is Q4 guidance, not Q3. Analysts are modeling Q4 revenue in the $38 to 42 billion range. If Micron guides in that range and signals continued HBM pricing strength into 2027, the June 23 sell-off will likely look like the classic pre-earnings dip in hindsight. If guidance disappoints or management sounds cautious on HBM volume trajectory, the reaction could be worse than the Korea-triggered drop.

Micron vs SanDisk Moat: HBM Advantage vs NAND Scale

Micron's competitive advantage is structural and rare. Making High-Bandwidth Memory at commercial scale requires a manufacturing process called chip-stacking, where multiple memory dies are physically bonded together using through-silicon vias (TSVs). This is technically complex, capital-intensive, and took years to develop. Globally, only three companies can do it: SK Hynix, Samsung, and Micron. Nvidia has certified all three for its Vera Rubin platform. No fourth player is close.

Beyond HBM, Micron is the only major HBM supplier headquartered in the US, which gives it geopolitical positioning no competitor can easily replicate. The company is building a $100 billion-plus semiconductor campus in Clay, New York with Bechtel as construction partner, backed by CHIPS Act funding. It opened India's first semiconductor assembly and test facility in Sanand, Gujarat in February 2026. This is a company building physical infrastructure on multiple continents as fast as it can.

SanDisk's moat is different but real. It comes from three sources: scale in NAND manufacturing through its joint venture with Kioxia (one of the most cost-efficient NAND production arrangements globally), technology node leadership in QLC and BiCS8 NAND that hyperscalers are now qualifying for enterprise SSD deployments, and the new business model contracts that convert spot-market NAND commodity sales into multi-year contracted revenue with financial guarantees. The $42 billion in contracted minimum revenue across five NBM agreements is unusual for any hardware company, and especially unusual for a NAND business.

The honest challenge to SanDisk's moat is cyclicality. HBM pricing is defended by manufacturing complexity. NAND pricing responds to supply additions faster, and Samsung's capacity expansion is a known variable for 2027. SanDisk has no DRAM or HBM to hedge against a NAND cycle turn. When NAND moves, the entire P&L moves with it.

Micron and SanDisk Analyst Ratings and Price Targets

Micron has a Strong Buy consensus across most major firms. Here are the key price targets as of June 22-23, 2026:

Goldman Sachs is the notable outlier with a Hold, a position that has cost clients meaningful upside in 2026 but reflects ongoing caution about cyclical peak-earnings risk and potential oversupply in 2027-2028 when new fab capacity comes online.

For SanDisk, the range of analyst targets is wider and the story is more unusual: several major firm targets now sit below where the stock was trading before the June 23 sell-off.

Morgan Stanley's Joseph Moore is worth unpacking separately. His Overweight rating stands, but his $1,750 target now sits 11% below the pre-sell-off close. He is not turning bearish. In a note following investor meetings with SanDisk management the week of June 21, Moore wrote that NAND supply and demand remain 'very tight' with no line of sight to balance. His target reflects through-cycle earnings at 23x, meaning he is anchoring on what the business earns across a full cycle rather than at peak. Cantor Fitzgerald's $2,900 target, by contrast, reflects a thesis that the NAND market has permanently re-rated and the trade is still in the 'mid-innings' of a new AI-driven paradigm.

Micron vs SanDisk Valuation: Which Stock Looks Cheaper?

Here is the most interesting part. On a forward earnings basis, Micron, the stock that rallied less, trades cheaper than SanDisk, the stock that rallied more. That is a meaningful observation for anyone trying to figure out where relative value sits.

At roughly 10.75x FY2027 consensus earnings, Micron is priced well below the semiconductor sector's median forward PE of approximately 36x and well below AI names like Nvidia that trade at 28 to 35x forward earnings. For a company with the world's only US-headquartered HBM franchise, 2026 production fully sold under contract, and FY2026 EPS growing at 651%, that multiple suggests the market is still pricing Micron like a cyclical commodity company rather than an AI infrastructure one. The TD Cowen, BofA, and Needham price targets all rely on essentially the same logic: memory is re-rating from commodity to infrastructure, and 10x forward earnings does not reflect that.

SanDisk's forward multiple is more complex. On consensus 2027 estimates of $105, the stock trades at roughly 18.7x. On Morgan Stanley's more bullish 2027 estimate of $149.68, the multiple compresses to 13x. Which figure is closer to reality depends on how long the NAND pricing cycle holds and how quickly the NBM contracted revenue ramps. If you use SNDK's own Q4 guidance midpoint of $31.50 EPS and annualize it, the implied run rate is roughly $126, which puts the stock at approximately 15.6x on a near-term annualized basis. These are not stretched multiples for a company growing the way SanDisk has grown.

As a framework for thinking about entry levels (not a recommendation, simply the math): for MU, a further dip to the $950 to $1,000 range would put the stock at roughly 9.7x to 10.2x FY2027 consensus EPS, which is close to the multiple at which value-oriented long-only funds have historically stepped in during memory dislocations. For SNDK, a correction to the $1,500 to $1,600 range would bring the stock to approximately 14 to 15x consensus FY2027 EPS, a level where the valuation story becomes considerably more defensible against a bearish scenario.

Neither of these represents a specific call on where the stocks go. They represent the price levels at which the forward earnings math starts to look genuinely compelling rather than already reflecting a lot of upside.

The Bear Case: Key Risks for Micron and SanDisk Investors

Any honest analysis of these two stocks has to engage with the downside scenarios. There are three legitimate ones, and they apply with different intensity to each company.

First, the supply response. Memory cycles are notoriously boom-bust. Periods of extreme supply shortages historically attract aggressive capacity additions that eventually overshoot demand. Micron is spending more than $25 billion in capex in FY2026. SK Hynix and Samsung are investing heavily too. Micron's own management has said new capacity from the New York campus will not come online in meaningful volumes until fiscal 2028 at the earliest. That timeline is coherent but not guaranteed. If all three major HBM manufacturers and multiple NAND producers expand simultaneously, a supply glut in 2027-2028 is possible.

Second, AI efficiency. The field learned in late 2025, with the emergence of highly efficient models trained at fraction of the expected compute cost, that algorithmic improvements can reduce hardware demand faster than infrastructure players expect. If frontier AI labs discover ways to run models with substantially less memory, demand projections get revised downward. This is the hardest scenario to quantify because it depends on the pace of improvements that nobody has consistently predicted accurately.

Third, the Chinese NAND competition. Chinese producers are scaling domestic NAND capacity aggressively. Their HBM capability remains years behind, but the gap in NAND technology narrows faster. Any significant reduction in the current supply-demand imbalance from Chinese NAND expansion would hurt SanDisk's pricing power harder than Micron's, since HBM pricing is structurally defended by manufacturing barriers that NAND does not have.

For existing holders, the question is whether these are known risks already reflected in the current multiples. At 10.75x FY2027 for MU and 18.7x FY2027 for SNDK on consensus estimates, there is an argument that cyclical discount is already baked into valuations, especially for Micron. For anyone considering fresh exposure, these scenarios represent the primary reasons for wanting a margin of safety on entry price.

Our Take On Which AI Memory Stock Looks More Defensible?

Both of these stocks have earned their returns. The supply-demand imbalance for AI memory is real, the multi-year contracted revenue is real, the margin expansion is real, and the hyperscaler capex underpinning the demand is real. These companies have structurally changed, not just cyclically benefited.

That said, no investment thesis is made stronger by ignoring what is already priced in. Both stocks are priced for a sustained AI infrastructure upcycle. Micron at $1,051 requires continued HBM execution, margins sustaining near record levels, and FY2027 EPS growing into $97 estimates that represent nearly 10x what the company earned in FY2024. SanDisk at $1,963 requires NAND pricing staying constrained well into 2027 and the new business model contracts converting into durable earnings at or above consensus.

For investors comparing the two on fundamentals at current prices: Micron looks more defensible. Its HBM oligopoly moat is harder to replicate than SanDisk's NAND position. Its forward earnings multiple is lower. Its diversified product base (HBM, conventional DRAM, and NAND) provides a hedge that a pure-play NAND company does not. A correction in NAND pricing does not sink Micron the way it would sink SanDisk. The Anthropic partnership announced June 22 adds a co-design dimension that moves Micron's customer relationship from vendor to infrastructure partner, which is a qualitative moat deepening.

SanDisk is the higher-upside, higher-risk version of the same AI storage bet. Its 613% YTD return reflects genuine re-rating, genuine earnings delivery (78.4% gross margins, $42B in contracted revenue), and some amount of pure-play premium and retail momentum. The business is real. The question is whether the current price leaves enough room for the cycle to be anything less than perfect.

The June 23 sell-off, triggered by a Korean market crash with no company-specific news, did not change the fundamental thesis for either company. What it did change is the entry price. Whether that dip is enough, or whether a deeper correction would offer a better risk-reward, depends on your own conviction in the earnings trajectory and your tolerance for the cyclical scenario. Micron's Q3 earnings, reporting tonight, will be the first real data point in answering that question.