Perhaps the most interesting part is not the target itself. Targets get revised all the time, and plenty of them age badly. What caught me was the claim that investors are still early in figuring out what this business is actually worth. I have found that memory stocks train you to distrust permanence. The cycle has burned enough people that a triple feels like a warning label. Still, the setup this time is odd enough that writing it off as the usual boom feels lazy.
Why Micron Stock Still Looks Misread After a Huge Run
Memory has always been the unglamorous cousin in semiconductors. Logic chips get the keynotes. Memory gets the inventory write-downs. For years the playbook was simple. Prices spike, producers add capacity, prices collapse, shareholders swear they will never touch the group again. Then the next shortage arrives and the same dance starts over. That muscle memory is why a stock that has already risen 266 percent in 2026 can still be called under-owned in spirit, even if the chart looks crowded.
The argument from the more bullish side is that the old reflex is now the mistake. After meetings with the company, one analyst concluded that the market is only beginning to assign a higher multiple, not finishing the job. I am not married to any single price target. A number like $3,000 is a scenario, not a promise. What I do take seriously is the mechanism underneath it. If demand for memory really does outrun supply through 2027 and 2028, the earnings power people are plugging into their models may still be too timid.
Memory is a lever for better AI performance. Models generate better results with more memory, run faster with more memory, and hold longer context windows with more memory.
Paraphrased from a recent analyst note after management meetings
That sentence is worth sitting with. It reframes memory from a commodity input into a performance knob. Once customers treat extra memory as a way to make the product better, not merely a cost to minimize, the bargaining position changes. Shortages stop being a temporary inconvenience and start looking like a feature of the buildout.
The Rally Already Happened, So Why Talk About Another Triple
Skeptics have a clean case. A stock that has climbed this far is not a secret. Consensus is already friendly. Roughly 46 of 50 covering analysts rate the shares a buy or a strong buy. When almost everyone agrees, the easy money is often gone. I have watched that movie in other chip names. The last buyers arrive just as the supply response shows up in the data.
The counter is uglier and more specific. The upside case is not that the stock is undiscovered. It is that the earnings base people are capitalizing is still being marked up. A higher multiple on a higher earnings number is how you get a path toward something close to a triple even after a huge year. That only works if the shortage lasts and if contracts lock in a chunk of the revenue before the cycle turns. Both of those are claims, not facts. They are also the claims that decide whether this is a trade or a hold.
Think of it like renting versus owning the upside. In a classic memory boom you rent the price spike. You hope to be out before the new fabs come online. In the version being sketched now, a large slice of revenue is meant to sit inside multi-year agreements with strategic customers. Those deals are not casual purchase orders. If they really are hard to cancel, the revenue line starts to look less like a weather report and more like a backlog. That is the shift that could pull in a different kind of investor, the sort who would never have owned a pure cyclical at any price.
What the New Price Target Is Actually Saying
A 43 percent hike in a target, from $2,100 to $3,000, is not a rounding error. It says the prior model was already bullish and still not bullish enough after the latest conversations. Implied upside of about 187 percent from a close near $1,045 is the kind of figure that makes portfolio managers check whether the analyst has lost the plot. Sometimes they have. Sometimes the plot really did change.
I read that revision as a statement about duration more than about next quarter. Near-term memory pricing can gap around on a single data point. A twelve-month target that far above the market is a bet that tight supply, richer mix, and stickier contracts survive long enough for the market to stop treating the earnings as peak and fleeting. If that bet is wrong, the target will look absurd in hindsight. If it is right, today’s price may still be discounting a cycle that ends sooner than the order book does.
Memory as the Quiet Lever Inside AI Systems
Most of the public conversation about artificial intelligence still orbits the processors that do the math. Fair enough. Those chips are astonishing, expensive, and scarce in their own way. But a model does not live only in the arithmetic. It lives in the data it can reach quickly. Add memory and the same model can keep a longer thread of context, retrieve more intermediate results, and avoid the clumsy workaround of constantly shuffling information back and forth. Cut memory and you get a fast engine attached to a tiny fuel tank.
That is why the shortage story has teeth. Training clusters and inference fleets both want more high-bandwidth memory per accelerator. Data centers want more conventional DRAM around the servers that feed those accelerators. Devices at the edge want enough local memory that they do not have to phone home for every prompt. None of this is theoretical. It shows up as customers asking for allocation, not for a discount.
In my experience, the market underprices a bottleneck until the bottleneck becomes a headline, then overprices it once the headline feels familiar. We may be in the awkward middle. The headline is familiar. The physical constraint is not finished. New capacity in memory does not appear because a press release says it will. Clean rooms, tools, yields, and packaging all take time. Even a motivated industry cannot wish a mature high-bandwidth line into existence in a quarter.
- More memory improves model quality, not just speed, because the system can hold more context.
- More memory shortens the path between compute and data, which is where a lot of real-world latency hides.
- More memory per accelerator raises the content of each server, which lifts dollar content even if unit growth cools.
- Tight supply into 2027 and 2028 is the claim that separates a spike from a multi-year earnings bridge.
Notice what is missing from that list. Nobody serious is arguing that memory stops being cyclical. The argument is narrower. The downcycle, if it comes, may start from a higher plateau, and a chunk of volume may already be spoken for. That is a different investment from “prices go up forever.”
High Bandwidth Memory and the Ordinary Chips Beside It
It helps to separate the glamorous stack from the workhorse. High bandwidth memory, the stacked product tied most tightly to AI accelerators, is the part everyone wants to talk about. It is harder to make, harder to package, and easier to sell out. Margins there can look nothing like the old commodity DRAM averages. That mix shift alone can re-rate a memory company if investors believe the mix sticks.
Then there is the rest of the portfolio. Standard DRAM still feeds servers, PCs, and phones. NAND still feeds storage. Those markets can be soft even while the AI slice is on fire, and they can snap back when PC and phone inventories finally clear. A company that sells both gets a blended story. The blend is a gift when the hot product carries the quarter. It is a headache when investors want a pure AI multiple and keep tripping over the legacy lines.
I have found the blended story more honest. Pure-play narratives are tidy and usually wrong. Micron is not a software firm with 80 percent incremental margins and no factories. It is a manufacturer with enormous capital needs, a handful of global peers, and customers who remember every price hike. The bull case has to survive that reality, not float above it.
Supply Cannot Be Summoned on a Conference Call
Every memory boom ends the same way in the textbooks. Producers earn abnormal returns, announce capacity, and collectively ruin the price. The textbooks are not useless. They are just slow. A fab is not a software deploy. Tool lead times, construction, qualification, and yield ramps mean that a decision taken today may not hit volume for a long stretch. If demand is still climbing while that capacity is only a plan, the shortage can outlast the skepticism.
Analysts who just met management think demand is likely to outstrip supply over the next two years. That is the whole ballgame for anyone underwriting a target miles above the quote. Two years is long enough for earnings to compound and for a new shareholder base to get comfortable. Two years is also short enough that a slip in demand, or a faster supply response from rivals, can wreck the math. I would not pretend the range of outcomes is narrow.
A rough way to hold the debate in your head: Price today reflects a boom the market already sees. A much higher target needs the boom to last. Contracts decide how much of that boom is pre-sold. Capacity decides when the pre-sold portion stops being special.
There is a human element here that models skip. Memory executives have been punished for overbuilding. The institutional scar tissue is real. A management team that has lived through a brutal downcycle does not add wafers for fun. They add them when customers will sign, when returns clear a hurdle, and when they think rivals will not flood the same node at the same time. That caution can extend a shortage. It can also mean they miss a window. Both risks sit on the table.
The Contract Shift That Could Change Who Owns the Stock
Here is the piece I keep coming back to. The company wants about half of revenue tied to long-term strategic customer agreements. Half is not a footnote. If those agreements are genuinely difficult to cancel, then a large part of the next few years is spoken for before the spot market has its usual mood swing. That is the reassurance aimed at investors who like growth but hate cliff edges.
Will customers actually honor that structure when the cycle turns? History says powerful buyers renegotiate when they can. A contract is only as strong as the alternative. If there are three hungry suppliers and a soft end market, paper promises get revisited. If there is one qualified source for a specific high-bandwidth stack and the customer’s own product launch depends on it, the paper holds. The mix of those two situations is what “strategic” has to mean. It cannot mean every purchase order with a longer date on it.
Still, even a partial shift matters. A business that used to live and die by weekly price checks starts to look a little more like an industrial supplier with a backlog. Multiples follow perception. Perception follows the first few quarters where guided revenue does not collapse the moment a headline turns sour. I suspect that is the journey the bullish note was describing. Not a sudden discovery of the company. A slow admission that the earnings might be less fictional than the old cycle trained people to assume.
If the contracts are real, the stock stops being a pure bet on next quarter’s price and starts being a bet on who already reserved the output.
How a Higher Multiple Sneaks Into the Model
Valuation is where these stories get sloppy. People say “it deserves a higher multiple” as if multiples were medals. A multiple is a compressed opinion about growth, duration, and risk. Memory names have traded at low multiples because duration was short and risk was high. Lift the duration, cut the perceived risk, and the same earnings stream is worth more. Lift the earnings stream at the same time and the price gap can look ridiculous on a spreadsheet.
That is how you get from a stock that has already tripled to a target that implies it could nearly triple again. You are not assuming the chart repeats out of momentum. You are assuming earnings power steps up and the discount rate the market applies to those earnings steps down. Both assumptions can fail. Earnings can miss if pricing rolls over. The multiple can compress if investors decide the contracts are marketing. I would rather see that tension written down than smoothed into a single confident adjective.
| Piece of the debate | What bulls emphasize | What skeptics emphasize |
| Recent move | Still early in the re-rating | 266 percent in one year is late |
| Demand | AI needs more memory to perform | AI spending can pause without warning |
| Supply | Tight through 2027 and 2028 | Capacity plans always arrive eventually |
| Contracts | Half of revenue targeted as strategic | Buyers renegotiate when leverage flips |
| Valuation | Multiple should rise with duration | Peak margins rarely get peak multiples for long |
| Street view | 46 of 50 analysts already positive | Crowded agreement is its own risk |
I like a table when the argument has two honest sides. This one does. Anyone selling you only the left column is pitching. Anyone selling you only the right column is recycling the last downturn. The useful work is deciding which column is more likely over the specific window you actually hold the shares.
Customers Are Not Buying Memory for Decoration
Follow the purchase back to the buyer’s problem. A cloud operator building an AI cluster is not shopping for memory the way a phone maker shopped for NAND in 2018. The operator is trying to ship a service that answers faster, holds more context, and costs less per useful token. Memory sits on that path. Skimp on it and the expensive accelerators idle while they wait for data. Overbuy it and the bill shocks the finance team. The equilibrium, for now, has been to overbuy relative to old server designs, because the alternative is a worse product.
That is a stronger demand signal than a consumer refresh cycle. Consumers can delay a phone. An infrastructure buyer racing peers for model quality has a harder time delaying the bill of materials. The risk is correlated. If several large buyers slow capital spending at once, the “strategic” queue shortens together. Concentration is the tax on this story. A few customers matter a lot. Their budgets are not independent of interest rates, power availability, or their own shareholders getting nervous about AI returns.
Power is the constraint I do not hear enough about in memory notes. You can have the chips and still lack the electricity and the cooling to turn them on. A data center that cannot get a grid connection does not need another pallet of modules this quarter. I have no clean way to model that from a memory income statement. I just know it can cap the upside without any change in the chip itself.
The Old Cycle Is Not Dead, It Is Just Wearing a New Coat
Let me be plain. I do not believe memory has been promoted out of the cycle club. Prices will fall again. Inventories will build again. Someone will add a fab at the wrong time. The question is amplitude and timing, not whether gravity still works. A business can be cyclical and still be a good investment if you buy the right part of the swing and if the midpoint of the swing has moved up.
The coat is the AI mix. When a rising share of bits sells into a use case that did not exist at scale five years ago, the midpoint can shift. High bandwidth products carry more revenue per wafer when yields cooperate. Long agreements can chop off the worst of the spot collapse. None of that repeals the cycle. It changes the floor people should argue about. Bulls think the floor is much higher than the last recession in memory pricing. Bears think the floor is the same old basement, just with a fancier staircase on the way down.
My own tilt, and it is a tilt rather than a creed, is that the floor is higher than the last cycle and lower than the current spot narrative. That middle view does not produce a neat triple. It does produce a reason to stay interested after a huge run, which is rarer than the internet makes it sound.
What a 50 Percent Contract Target Really Changes
Numbers like 50 percent invite false precision. Half of revenue under strategic agreements sounds like a switch. In practice it will be a slope. Some agreements will be firm. Some will be firm until they are not. Some will cover volume and leave price floating. Some will cover both and include outs that lawyers understand better than shareholders. The headline percentage is a direction. The footnotes are the investment.
Why push for that structure at all? Because the company knows its own history. Volatile pricing scares off the exact shareholders who could support a higher multiple. Lock in a strategic book and you can talk to those shareholders without flinching every time a spot index ticks down. There is a self-serving angle, of course. Guaranteed volume also justifies capacity spending. Customers who sign are helping fund the supply they later complain is tight. That loop can be healthy. It can also be how an industry talks itself into the next glut.
- Watch the share of revenue described as strategic, not just the ambition to reach half.
- Listen for whether price, volume, or both are actually committed.
- Notice cancellation language, even when it is buried.
- Compare the contract story with inventory days and with what peers are building.
- Treat a single quarter of stable guide as a clue, not a conversion.
That checklist is deliberately dull. Dull is useful when a stock has a story this loud. The loud part is the AI lever. The dull part is whether the revenue is cancellable. I would rather be dull and still invested than excited and surprised.
Peers, Pricing Power, and the Small Club Problem
Memory is an oligopoly in practice. A few producers matter. That structure cuts both ways. Coordination is not formal, yet behavior rhymes. When all of them see the same shortage, none of them is in a hurry to give away price. When all of them see the same empty fab slot, all of them are tempted to fill it. Investors who model one company in isolation miss the game.
Pricing power in a shortage is real and temporary. It feels permanent right up until the second source qualifies. For the highest-end stacks, qualification is slow, which extends the temporary period. For mainstream DRAM and NAND, qualification is faster, which shortens it. A blended company gets both clocks running at once. The reported gross margin is an average of a privileged business and a contested one. Averages hide the thing you actually own.
I have found it helpful to ask a blunt question. If the AI slice stopped growing for two quarters, would the rest of the portfolio carry the valuation, or would the stock give back a year of gains in a month? If the honest answer is the second one, position size matters more than the beauty of the thesis. A good story with a violent left tail is not a core holding just because the right tail is a triple.
Capital Spending Is the Bill That Arrives Later
Fabs are not free, and memory fabs are especially not free. The same tightness that lifts margins also pressures management to spend. Shareholders cheer the margins and then grumble about the spending, sometimes in the same paragraph. The sequence is predictable. Underinvest, shortage, overearn, overinvest, glut, underinvest. Breaking that sequence is the entire strategic point of the contract push. Spend against signed demand, not against hope.
Hope is still in the model. Any multi-year supply plan assumes customers will be there when the tools are. If power constraints or a pause in AI capital expenditure arrive first, the spending looks early. Early spending is how return on capital dies. I do not have a proprietary view on the next capex guide. I do have a bias that the market will forgive spending while the shortage is obvious and punish it the moment the shortage looks less obvious. That lag is where holders get hurt.
A simple filter before chasing upside: shortage evidence + contract quality - capex optimism = how much of the target you are willing to underwrite.Filters like that are not science. They keep me from treating a price target as a destination. The $3,000 figure is someone else’s destination. Mine has to survive a cut to the multiple and a cut to the margin at the same time, because those two cuts tend to travel together.
Geopolitics Sits in the Background and Sometimes in the P and L
You cannot talk about a U.S.-based memory producer without talking about where the wafers, the tools, and the customers actually are. Export rules, subsidy races, and regional diversification all leak into cost and into who is allowed to buy the best parts. A rule change can remove a customer or protect a price. Neither shows up cleanly in a demand model built from AI server counts.
I am wary of turning this into a policy essay. The practical point is narrower. Supply chains that look tight can be made tighter by regulation, and demand that looks endless can be redirected by the same pen. Anyone underwriting two more years of shortage should at least ask which part of that shortage is physical and which part is political. Physical shortages resolve with tools and time. Political ones resolve on a different calendar, sometimes overnight.
There is also the quieter operational risk. Concentrated manufacturing footprints are efficient until they are not. A disruption at a single site can move a global price. Investors call that an externality until it is their quarter. Position sizing, again, is the adult response. Not a slogan about resilience.
What the Broad Analyst Agreement Does and Does Not Mean
Forty-six buys out of fifty is a fact about the street, not a fact about the future. Agreement can mean the evidence is strong. It can also mean the career risk of being the skeptic feels higher than the career risk of being wrong with the group. I have sat through both versions. The useful tell is whether estimates are still rising. A crowd that is positive and still lifting numbers is different from a crowd that is positive and done.
The latest target hike suggests at least one desk is not done. A move from $2,100 to $3,000 after a face-to-face with management is a claim that the public story lagged the private one. Maybe. Management meetings are also where optimism is most contagious. People who fly to headquarters rarely fly home to cut numbers on the same day. I discount the theater and keep the mechanism. If the mechanism is longer context windows, tight supply, and harder-to-cancel agreements, the meeting is just the delivery system.
Would I need the street to be split before I cared? Not really. Splits are comforting and often late. What I need is a reason the positive case survives a bad month. The contract target is that reason, if it is real. The AI performance lever is that reason, if customers keep paying up for it. Consensus is just the crowd standing on those reasons.
How I Would Frame the Upside Without Falling in Love With It
Let us put the triple in plain language. From a little over a thousand dollars a share to three thousand is not a gentle re-rating. It assumes the market is wrong by a wide margin on either earnings, duration, or both. Wide-margin bets are allowed. They should not be dressed up as base cases. I would label the $3,000 path as the version where supply stays tight, mix stays rich, and half the revenue really does behave like a backlog. Label the current price as the version where the market believes some of that and not all of it.
Between those poles there is a lot of room to be right and still not make a triple. Earnings can double and the multiple can stay grumpy. The multiple can expand and earnings can stall. A holder who needs the extreme path to justify the position is not investing. They are hoping the note was a script. Notes are not scripts.
There is a personal tell I trust more than targets. When I notice myself repeating a catchy line, I slow down. If you don’t buy it, they will. It is a good line. It is also a sales line. The “they” is other investors, or customers, or both. Other investors can be early or they can be the greater fool. Customers can be locking supply or they can be double-ordering. The line does not distinguish. The filings and the pricing data eventually will.
Scenarios Worth Writing Down Before the Next Gap
Memory stocks gap. They do not drift politely while you finish your homework. So the homework has to exist before the gap. I keep three sketches, none of them sacred.
In the tight-and-sticky sketch, AI server builds stay strong, high bandwidth yields improve, and strategic agreements cover a rising share of sales through 2027. Spot prices cool without collapsing. The multiple inches up because the earnings look repeatable. This is the sketch that can approach the aggressive targets. It needs cooperation from customers, from tools, and from rivals who do not flood the node.
In the classic-cycle sketch, pricing stays hot for a few more quarters, capex plans multiply, and a large buyer pauses. Contracts get “adjusted.” The stock gives back a large piece of the 2026 move even if the company is still nicely profitable. This sketch has the longer historical record. Ignoring it because the current quarter feels different is how people donate money to the cycle.
In the muddle sketch, which is the one I weight most, AI demand stays real, supply stays tight longer than bears expect, and the multiple never fully leaves the cyclical neighborhood. Shareholders who bought the panic make a lot of money. Shareholders who buy every uptick make some, then sit through a drawdown that feels unfair. The business is better than the last cycle. The stock still behaves like a stock with operating leverage and a mood.
Weight those however you like. The point of writing them down is to stop treating one analyst path as the map. A map with one road is a brochure.
Signals That Would Make the Bull Case Stronger
I do not need a new slogan. I need evidence that the slogan is becoming an operating fact. A few signals would move me.
- Strategic agreements described with enough detail to judge duration and cancellation, not just a percentage goal.
- Gross margin that stays elevated even when a non-AI end market is soft, which would suggest mix is doing real work.
- Inventory that does not balloon at customers while the company talks about scarcity.
- Capex that is explained against signed demand rather than against a total addressable market slide.
- Peer commentary that rhymes on tightness instead of quietly guiding to share gains at lower price.
None of those require a price target. They require patience with transcripts, which is less fun than a round number and more useful. If the signals show up, the idea that investors are early stops sounding like a pep talk. If they do not, the round number is just a round number.
Signals That Would Make Me Step Aside
The other list is shorter and meaner. A meaningful customer pushing out AI server orders. A sudden improvement in high bandwidth availability from a second source. Inventory days rising while management still uses shortage language. A contract percentage that stalls because buyers will only sign flexible deals. Any one of those does not kill the franchise. Together they kill the duration assumption, and the duration assumption is what the extra multiple needs.
I would also step aside, or at least cut size, if the stock started trading as if $3,000 were the base case rather than the stretch case. Markets do that. They take a bull note and install it as destiny for a few weeks. Destiny is a bad entry. The business can be excellent and the entry can still be careless. Those are different skills, and only one of them shows up in a product teardown.
Position Size, Time Horizon, and the Temptation to Be Clever
A stock that can plausibly be worth a lot more and can also drop hard is a sizing problem before it is a forecasting problem. I would rather own a smaller piece of a violent compounder than explain to myself why I “knew” the cycle and still rode it down with a full weight. Cleverness is the enemy here. The clever trade is to rent the shortage and leave before the contracts get tested. The clever trade is also how people miss the entire re-rating if the contracts hold.
Time horizon sorts the cleverness. If your horizon is a quarter, you are trading pricing data and you should say so. If your horizon is the two-year window where supply is supposed to lag, you are underwriting duration and you should be willing to sit through ugly weeks. Mixing those horizons is how a long-term thesis becomes a short-term panic. I have done that mix. It is expensive tuition.
There is no trophy for suffering, either. A thesis that needs two years does not require you to ignore a broken mechanism. If the agreements prove soft, the two-year story is over even if the calendar is not. Holding out of pride is not patience. It is a different mistake with a better vocabulary.
What This Means If You Already Own It
Holders who bought much lower are in a pleasant and dangerous spot. Pleasant because the move has already paid for a lot of future error. Dangerous because the gain itself becomes an argument to stay, which is not an argument. The question is whether the next buyer, the one looking at $1,045 rather than at your cost basis, is being offered a decent proposition. If the answer is yes, holding is coherent. If the answer is only yes in the extreme sketch, trimming is coherent too.
I do not think existing holders need to apologize for the rally. A 266 percent year is not a bubble by definition. It can be the market catching up to a profit pool that was mis-modeled. It can also be the market extrapolating a profit pool that is about to be competed away. Both have happened in this industry, sometimes in the same decade. The holder’s edge, if there is one, is the willingness to reread the contract story every quarter instead of rereading their own cost basis.
What This Means If You Are Still Outside
Being late is not the same as being wrong. A stock can be up fourfold in a year and still be mispriced if earnings are about to be up by more than the market has capitalized. That sentence should make you uncomfortable. It is also the only sentence that justifies looking at a fresh buy after a move like this. You are not buying the chart. You are buying the claim that the chart has not finished discounting the shortage.
Outside money should demand a discount to the story, not a tribute to it. Chasing a target because the target is round is how outside money becomes exit liquidity. Building a position only if the signals on contracts and supply keep improving is how outside money gets to participate without pretending the easy part remains. The easy part was lower. It is gone. What may remain is the harder part, which is sometimes where the multiple actually changes.
If that sounds like a hedge, good. A hedge is appropriate. The phrase about other people buying if you do not is an invitation to urgency. Urgency is a trading emotion. The underlying business, if the bulls are right, does not require you to decide before lunch.
A Few Misreadings I Keep Seeing
One misreading is that a higher target means the risk fell. It often means the estimated reward rose, which is not the same thing. Another is that analyst agreement is a substitute for customer evidence. A third is that AI demand is a single switch. It is a cluster of budgets, power contracts, model releases, and internal return hurdles. Switches are easier to model. Clusters are what you actually get.
A fourth misreading, and the one I am most prone to, is treating a memorable sentence as analysis. Lines travel because they are short. Businesses pay you, or do not, based on wafers, yields, and signed volume. I can like the dare and still insist on the wafers.
A fifth is the idea that memory is suddenly a software-like annuity. It is not. Even with better contracts it remains a factory business selling into concentrated customers in a politicized supply chain. Annuity language is how people justify annuity multiples. I would not pay an annuity multiple for a business that can still have a bad pricing year. I might pay a higher cyclical multiple. The difference is the whole valuation argument, hiding in a vocabulary choice.
Putting the Pieces Back on One Page
So where does that leave the claim that Micron stock could nearly triple from here? It leaves it as a conditional, not a headline destiny. The condition is a shortage that really does run through the next two years, a product mix that keeps memory acting like a performance lever rather than a commodity filler, and agreements that cover something close to half of revenue without falling apart at the first sign of buyer leverage. Stack those and a much higher price is not absurd. Drop any one of them and the path back toward the old multiple is short.
The stock has already done the part that makes people late. Up 266 percent in the year, up more than 460 percent across twelve months, widely rated a buy, and still being handed a target nearly triple the recent close. That combination is rare enough to deserve a careful look and common enough in late cycles to deserve suspicion. I can hold both feelings at once. Markets do not require me to pick a personality.
What I will not do is pretend the move so far used up the fundamental question. The fundamental question is whether AI systems keep paying for more memory because the product is worse without it, and whether the industry can resist flooding that demand before the contracts have done their job. Everything else, including a very large price target, is commentary on that question.
The chart already told you what happened. The contracts, the yields, and the next wave of capacity will tell you whether the rerate was early or late.
If you are still reading, you probably do not need another adjective about artificial intelligence. You need a way to stay honest while a popular stock gets a more popular target. Mine is simple enough to reuse. Separate the performance lever from the cycle. Separate the contract ambition from the contract terms. Separate the upside sketch from the position you can actually live with. Do that, and a wild target becomes a scenario you can decline, accept in part, or build around. Skip that, and you are just another person who heard that someone else would buy it if you did not.
I still do not know which sketch wins. I do know the debate is no longer about whether memory matters to AI. It matters. The debate is about how long that fact stays scarce, and what a scarce fact is worth once everyone can see it. That is a less catchy conclusion than a triple. It is the one I trust more on a Wednesday morning, after the note has circulated and before the next inventory number has a chance to argue back.
]]>
