The RTX 5090 launched at $1,999. It is now reportedly climbing to $3,843.
That is not a typo, and it is not a scalper's markup. That is the baseline, board-partner cards with fancy cooling on top of it.
Your first instinct, staring at that number, is the correct human instinct and the wrong strategic one: I'll just wait.
Wait for the shortage to pass. Wait for supply to catch up. Wait for the January sale that always comes. This piece exists to talk you out of that plan, because the math forbids it until at least 2028.
Last week we flagged Samsung's pricing caution around the S26 FE as an open question. This week Samsung answered it, and the answer is worse than the question implied: the memory crunch is now formally forecast to run through 2028.
Here is the sentence that should reorganize your buying calendar. "So we believe it will be unlikely to see any significant increase in incremental supply through 2028," said Jaejune Kim, Samsung's EVP of Memory, on the company's Q2 earnings call.
Not "prices stay high." Supply barely grows. Those are different animals, and the second one is the one that eats your savings plan.
Not 'prices stay high.' Supply barely grows. Those are different animals, and the second one is the one that eats your savings plan.
Understand why waiting normally works. Memory is a commodity that runs on a boom-bust cycle: prices spike, everyone builds fabs, new supply floods in, prices crater. Patience has beaten the memory market for decades.
That cycle just snapped. The buyer setting prices now is not you.
It is AI data centers. Model training was already devouring high-bandwidth memory; Samsung now says agentic AI has piled on top, driving what it calls an "unprecedented rise" in demand for AI servers.
The result shows up in Samsung's own ledger. Q2 operating profit reportedly jumped more than nineteenfold year-on-year, to the equivalent of $62.4 billion. Its memory business is printing records.
Its mobile division, meanwhile, reportedly posted a 700 billion won quarterly loss, dragged down by the very memory prices its other arm is celebrating. When one company's left hand is getting rich off what's bleeding its right hand, that tells you where the priorities sit.
And Samsung has made those priorities explicit. It plans to place roughly 60 to 70 percent of its memory capacity under long-term contracts, prioritizing customers "who can guarantee committed future captive demand."
In plain terms: sign up for years of purchases and you move to the front. As The Register put it, everyone else can take a number and get in line.
You, the person buying one laptop, are the back of that line. Consumer gear is now structurally the lowest-priority buyer in the entire memory supply chain.
So why won't a new supplier swoop in and undercut? Because building a memory fab is one of the slowest acts in industry: permits, ultra-pure water systems, lithography tools that take months to tune. Fab-to-wafer lead time exceeds three years.
The money is moving, to be clear. South Korea reportedly announced a $576 billion push led by SK Hynix and Samsung; Micron reportedly committed up to $3 billion to shore up US supply. None of it lands soon.
Analysts at TrendForce reportedly expect several DRAM lines to start in 2027, but say meaningful output won't materialize until 2028. IDC reportedly puts relief in the same window.
And notice who competes for the same scarce capacity even at the top of the stack. Nvidia and AMD both fight for TSMC's manufacturing slots, which is why the expectation is that AMD follows Nvidia's price increases rather than undercutting them. Nobody in this market has an incentive to start a price war.
Here is the darkest part, the reason "wait for relief" fails even in 2028. The relief, when it comes, is a bust engineered to save the makers, not you.
Memory vendors finance these giant fabs on the boom. They know that when new capacity switches on it can flood the market and collapse prices, and they are betting AI demand holds long enough to absorb it. If it does, they profit. If AI demand stumbles just as the fabs ramp, they face what's been called a bust to end all busts.
Either way, the mechanism that eventually cracks prices is a wholesale glut between memory makers and AI buyers. There is no promise that a data-center price collapse ever reaches the sticker on a mid-tier phone.
So let's turn this into a buy-timing call, category by category.
GPUs: buy or hold now. If you own a capable card, this is the year to make it last, full stop. If you genuinely need one, the reported Chinese-market rates are the writing on the wall: the RTX 5070 Ti reportedly heading toward $1,408, nearly double its $749 launch; the 5070 toward $973; the 5060 Ti near $783 from $429. Those are Nvidia's baseline numbers, before partner markups. There is no version of 2027 where these get cheaper.
Tablets and mid-tier phones: upgrade before the 2027 peak. Samsung has confirmed the Galaxy Tab S12 and Galaxy S26 FE for later this year and openly warned rising component costs could push pricing up. The S26 FE is rumored to get a more premium, flagship-level redesign, exactly the kind of memory-heavy spec bump that gets expensive fast in this environment. If one of these is your next device, buying into the 2026 launch beats buying into the 2027 crunch.
The tell is already on the shelf. The Galaxy Z Flip 8 arrived with a largely unchanged spec sheet and a $100 higher starting price. That is the shape of every upgrade to come: pay more, get the same, or pay the same and quietly lose memory and storage headroom.
Laptops: spec up now, or spec down later. If you buy this year, max the RAM and storage while it's the cheapest it will be for years, because those are the exact components under pressure. If you wait, expect to pay more for less, or to see manufacturers strip capability out to hold a price. Samsung is reportedly readying a budget SSD that removes its onboard DRAM entirely, borrowing system memory to stay cheap. That is the future of "affordable": affordable by subtraction.
The through-line across all of it: "wait for prices to drop" is the one strategy this market forbids. Supply barely grows through 2028. The lowest-priority buyer is you. And the eventual relief is a wholesale event that may never trickle down to a checkout page.
Buy the memory-heavy device you actually need before 2027. Not the one you might want, not a speculative upgrade, but the real replacement you can already see coming.
Because the reckoning here has only been deferred, not cancelled, and it was never scheduled to land in your favor.