Why Is RAM So Expensive in 2026? The Shortage Is a Choice

RAM is expensive in 2026 because the companies that make it earn more by keeping it scarce than by making more of it, and they are choosing scarcity. On June 24, 2026, Micron reported the most profitable quarter in its history: record revenue of $41.46 billion, up from $9.30 billion a year earlier, at gross margins near 85%. That is what the shortage looks like from the other side of the till. Prices have roughly tripled since late 2025, the three big makers have locked up supply through 2026 and beyond, and they have told investors they will not flood the market to fix it. This is not a supply accident waiting to clear. It is the most lucrative thing to happen to the memory business in decades, and that is exactly why cheap RAM is not coming back soon.

Micron just had its most profitable quarter ever

Start with the scoreboard, because it explains the behavior. On June 24, 2026, Micron reported record fiscal Q3 2026 revenue of $41.46 billion, against $9.30 billion in the same quarter a year earlier, with GAAP net income of $28.24 billion and a gross margin near 85%. The company guided the next quarter higher still. CEO Sanjay Mehrotra called it proof of “the strategic value of memory in the AI era.”

It is not just Micron. SK Hynix, the leading maker of HBM, posted a 72% operating margin in Q1 2026 with revenue up 198% year on year, and told investors plainly that “customer demand exceeds supply capacity.” Margins like these are unheard of in a commodity business. They are the financial signature of a shortage, and they are the reason the people who could end the shortage are in no hurry to.

Why is RAM so expensive? Because scarcity pays

The standard explanation is that AI demand created a shortage. True, but incomplete. A shortage that lasts is a shortage someone has decided not to fix, and the makers have decided.

Two of the three, Samsung and SK Hynix, told investors in late 2025 they would not pursue aggressive capacity expansion, even as prices ran. Together they control roughly 70% of the DRAM market, so that restraint sets the price for everyone. Micron, on the same theme, said it could fill only 55% to 60% of its core customers’ demand and is steering output toward higher-margin parts rather than racing to add commodity supply. All three now ration what they make through allocation-only frameworks that put hyperscalers and big PC makers first, which keeps the open market starved.

This is not a conspiracy. It is discipline, and it is rational. The makers were badly burned in the 2023 glut, when oversupply crushed prices and pushed them into heavy losses. The lesson they took away was that flooding the market destroys profits, so this time they are holding the line. When the same three companies that caused the 2023 crash by overbuilding now refuse to overbuild, “shortage” and “strategy” stop being different words.

How AI eats your RAM: the HBM crowd-out

The mechanism that makes the discipline possible is high-bandwidth memory, the stacked DRAM that sits next to a GPU in an AI server. It is where the money is and where the capacity goes.

HBM is expensive to make and lucrative to sell. By SemiAnalysis estimates cited in IEEE Spectrum, it costs about three times as much as ordinary memory and makes up half or more of the cost of a packaged GPU. Crucially, it is far more wafer-hungry: Micron has described HBM consuming roughly three units of DDR5 capacity for every unit of HBM produced. So each wafer a maker points at AI is effectively three wafers’ worth of ordinary RAM that never gets made. With the entire 2026 HBM supply already booked and the HBM market forecast to nearly triple by 2028, the pull on capacity is structural, not seasonal.

How high have prices actually gone?

Far enough to reset what normal looks like. DRAM prices rose 80% to 90% in the first quarter of 2026 alone, per Counterpoint Research data reported by IEEE Spectrum, one of the steepest single quarters on record. At retail, a 32GB DDR5 kit that sold for around $100 a year ago now runs about $375, per Tom’s Hardware tracking. The research firm Gartner has gone further, forecasting a combined DRAM and SSD price rise on the order of 130% across 2026, which it expects to push average PC prices up around 17%.

The increases have not been a one-off spike that snaps back, either. They stacked quarter on quarter through the first half of 2026, which is the pattern you get from withheld supply rather than a temporary demand shock. For the current week-to-week numbers, this post pairs with a live DRAM price tracker that follows the spot market from primary sources.

When will RAM prices drop?

Not in 2026, and probably not in 2027. The most-quoted forecast is the bluntest: at a February 2026 conference, Intel CEO Lip-Bu Tan said the memory industry told him “there’s no relief until 2028,” as reported by Bloomberg. Micron’s own commentary points the same way, with demand described as tight well beyond calendar 2027.

The arithmetic backs the pessimism. A new memory fab takes roughly 18 to 24 months to build and longer to reach good yields, and Micron’s new Idaho fab is not expected to reach volume until 2027. So even a crash program started today lands supply in 2027 or 2028. Add the makers’ deliberate restraint on top of the build time, and relief is slower still: the capacity only arrives when they decide the profits from scarcity no longer beat the profits from volume. The honest answer to “when does RAM get cheap again” is a range, not a date: elevated through 2026, maybe easing in 2027, normalizing toward 2028, and even then likely settling above the pre-AI baseline.

If you are deciding whether to buy now, the practical read is simple. Prices are more likely to rise than fall over the next several months, so buy what you actually need and skip the speculative stockpiling. For the full timeline, the competing forecasts, and a buy-now-or-wait call by buyer type, see when will RAM prices drop.

What it costs you, and what it costs AI

The bill lands in two places. First, on consumers and PC buyers. The research firm IDC expects the 2026 PC market to contract between 4.9% and 8.9% and smartphone shipments to fall 2.9% to 5.2%, with average selling prices up 3% to 8% as memory costs pass through. PC makers including Dell, Lenovo, HP, Acer, and ASUS have warned of price increases in the 15% to 20% range, per Tom’s Hardware reporting, and where they cannot raise prices they ship less memory for the same money. The rational consumer move when storage gets this expensive is to keep existing machines longer rather than buy into the spike, and to recover the value in old kit rather than bin it: directories that route retired hard drives to local recyclers keep the magnets, aluminium, and rare-earth content in circulation instead of sending it to landfill.

Second, and less obviously, the squeeze raises the cost of AI itself. Memory is one of the largest line items in a server, so pricier DRAM and HBM lift the cost of every GPU box and, with it, the cost of running inference. The demand that created the shortage now pays more to feed itself. That makes the economics of decentralized GPU capacity versus the big clouds and of self-hosting models on your own hardware more interesting, and it sharpens the open question behind the AI productivity paradox: whether the build-out earns back what it costs. For the model-by-model running costs, the cost-per-task calculator keeps the figures current.

Is this the new normal?

For a while, yes. The combination of disciplined makers and structural HBM demand means the baseline price of memory has reset higher, and even the eventual easing is unlikely to return prices to the pre-AI floor. The makers have learned that scarcity is more profitable than scale, and nothing about the AI build-out is telling them to change their minds.

The one force that ends it is the one that always ends memory cycles: the makers overbuild, supply overshoots demand, and the glut crashes the price, as it did in 2019 and again in 2023. That reckoning will come, because it always does. But it comes on the makers’ schedule now, not the market’s, and they are being paid extraordinarily well to delay it. Until the math flips, expensive RAM is the new normal, not the anomaly.

Sources

• Micron Technology (2026). Reports Results for the Third Quarter of Fiscal 2026 (SEC Form 8-K exhibit, June 24 2026). sec.gov
• SK Hynix (2026). Q1 2026 Business Results. news.skhynix.com
• IEEE Spectrum (2026). AI Boom Fuels DRAM Shortage and Price Surge (citing Counterpoint Research and SemiAnalysis). spectrum.ieee.org
• Sourceability (2026). Tracking memory price increases across the last several quarters (citing TrendForce, Counterpoint, and Micron earnings on capacity discipline and the HBM consumption ratio). sourceability.com
• Bloomberg (2026). Intel CEO Says There’s ‘No Relief’ on Memory Shortage Until 2028. bloomberg.com
• IDC (2025). Global Memory Shortage Crisis: Market Analysis and the Potential Impact on the Smartphone and PC Markets in 2026. idc.com
• Tom’s Hardware (2026). RAM price tracking 2026 and Framework warns of even more rising RAM and SSD prices through 2026. tomshardware.com
• Gartner (2026). Forecast of a combined DRAM and SSD price rise on the order of 130% across 2026, as reported in June 2026 (secondary). Treated as an attributed forecast.