Aug. 26, 2026

When does the RAM shortage end?

The question, verbatim from Hacker News: "Is there anything in the works or planned that suggests the RAMpocalypse will end anytime soon? e.g. new fabs being built." There are new fabs. Almost none of them produce anything before 2028. Here is what the people paid to forecast this actually say, name by name.

The forecasters mostly agree, which is itself unusual. The modal answer: DRAM stays tight through 2027; relief begins somewhere between late 2027 and 2028; a minority of voices — including the chairman of the group that owns SK hynix — say 2030. Counterpoint Research puts the first real relief in late 2027. UBS models supply and demand crossing around Q2 2028. TrendForce forecasts contract prices rising every quarter through the second half of 2027. J.P. Morgan says “multiple years” and declines to name an end date. One credentialed dissenter says a year. Nobody with a fab says sooner.

How severe is the RAM shortage in 2026?

The numbers below come from different kinds of series, and mixing them up is how bad headlines get written. Retail kit prices are what a person pays at checkout. Contract prices are what PC and server makers negotiate quarterly. Bank estimates splice both with forecasts. Labeled accordingly:

A retail kit that quintupled is not the same fact as a contract index that doubled in a quarter, and neither is a bank’s three-year cumulative figure that includes months that have not happened. All three point the same direction.

Why is there a RAM shortage?

Four mechanisms, all documented, none quickly reversible.

High-bandwidth memory is eating the wafer supply. HBM consumes roughly three times the wafer capacity per bit of ordinary DRAM, per IEEE Spectrum, so every production line converted to feed AI accelerators removes more conventional supply than it appears to. On top of that, OpenAI’s Stargate project signed letters of intent with Samsung and SK hynix for a reported ~900,000 wafers a month — reported as roughly 40% of global DRAM output. That figure is secondhand and the LOIs are not purchase contracts, a point that matters later in this piece.

The contracts reset at the worst moment. Fixed-price long-term supply agreements — including Apple’s — expired at the end of 2025, and the replacements are three-to-five year agreements with flexible pricing, per Igor’s Lab. The buyers who had been insulated from spot prices no longer are.

And the makers are not racing to fix it. TrendForce puts 2026 capex up just 14% for DRAM and 5% for NAND, aimed at process migration rather than raw bit growth. Memory makers remember 2019 and 2023, when overbuilding cratered prices. Scarcity is profitable.

When will new memory fabs start production?

New fabs are being built — the Hacker News commenter’s question has a literal yes. The dates are the problem. StorageReview’s summary of the Micron and SK hynix announcements is the frame to keep: almost nothing lands before 2028.

The one supplier adding meaningful conventional-DRAM bits before 2028 is China’s CXMT, which SemiAnalysis reports is running about 300,000 wafers a month now and heading to roughly 420,000 by the end of 2027 — around 17% of global DRAM. Its output is reportedly booked through 2027 anyway.

RAM shortage forecasts by source

Every date below belongs to the source named on its row. None of them is our forecast.

Kyung is the outlier worth taking seriously — he ran the world’s largest memory maker — and also a man no longer accountable for a fab’s utilization rate. Everyone currently running one says 2028 or later.

Could the RAM shortage end sooner?

There is a documented skeptical scenario, and it runs through demand rather than supply. Analysis carried on Yahoo Finance sketches a potential 2028-29 oversupply if AI capital spending slows: every fab in the table above arrives at once into a market that stopped absorbing it. Storage analyst Tom Coughlin has made the boom-bust argument for years — memory has never had a shortage that did not end in a glut.

Two facts feed that scenario. The Stargate letters of intent that anchor the demand story reportedly carry no purchase obligation; if AI budgets tighten, 40% of projected DRAM demand is a handshake. And demand destruction is already measurable at the consumer end: Gartner and IDC forecast PC shipments falling 10-11% and phone shipments 8-9% in 2026 because memory made the products too expensive. Shortages cure themselves from both directions; the argument is only about when.

Should Mac buyers wait for RAM prices to fall?

The forecasts do not support waiting a few months for broad price relief. Most put the first easing in late 2027 or 2028, and none guarantees that lower component prices will immediately reach retail Macs.

Apple lives inside this story, not beside it. Its fixed-price memory contracts were among those that expired at the end of 2025, its June repricing moved the whole Mac line up 14-33%, and the $899 Mac mini is what a memory shortage looks like when it reaches a product page. If the forecasters above are right, the machines shipping before Sept. 22 were priced for a shortage that runs at least another year — and memory is most of what you are buying in any machine meant to run models. We are not going to out-forecast TrendForce and J.P. Morgan; we will keep recording the prices.

About the data: The price series in this piece measure different things — retail kit prices, contract price indices and bank cumulative estimates — and are labeled in place; they are not interchangeable. Every forecast is attributed to its named source and none of them is ours. Secondhand figures, including the reported ~40% Stargate share of DRAM wafer supply and the Kearney 2030 estimate, are marked as reported by the outlets that carried them.