The global memory shortage has reached the MacBook Air, and that matters well beyond people shopping for a laptop. Bloomberg's Mark Gurman reported on August 2 that ship dates on Apple's own store have slipped to between two and six weeks, with the higher RAM configurations waiting longest, on a machine Apple refreshed only in March. Apple had already raised the base price by two hundred dollars in June. The cause is not an Apple problem. Samsung, SK Hynix and Micron are steering wafer capacity toward high bandwidth memory for AI accelerators, and every buyer of ordinary DRAM is now standing behind that queue.
The short answer
Bloomberg's Mark Gurman reported on August 2 that the M5 MacBook Air is short of stock, with ship dates on Apple's own store running two to six weeks and the higher RAM configurations waiting longest. Apple had already lifted the base price from 1,099 to 1,299 dollars in June, and has been sourcing memory from Chinese suppliers and delaying its back to school promotion. The Mac mini and Mac Studio were hit first. The cause is industry wide: DRAM makers are steering capacity toward high bandwidth memory for AI accelerators, and TrendForce expects server DRAM contract prices to rise thirteen to eighteen percent quarter on quarter in the third quarter of 2026.
We have been reading about the memory shortage as a market story for most of a year, in the way one reads about commodity prices: real, but happening to someone else. It stops feeling abstract when the company with the deepest supply contracts in consumer hardware cannot keep its most ordinary laptop in stock.
What was reported
Bloomberg's Mark Gurman wrote on August 2 that the M5 MacBook Air, in both the thirteen and fifteen inch sizes, is in short supply. Apple's own United States store is quoting two to six week delivery estimates, and the configurations with more RAM are the ones waiting longest.
That last detail is the whole story in one line. If assembly capacity or shipping were the constraint, every configuration would slip together. When the delay scales with how much memory you asked for, the constraint is memory.
Gurman cited retail sources describing incoming shipments as more constrained than they can recall, on a machine refreshed as recently as March. Apple has already raised the base price from 1,099 to 1,299 dollars, a two hundred dollar increase applied in June, and it still cannot meet demand. It has also been sourcing memory from Chinese suppliers, pushed its usual back to school promotion out of June, and started attaching availability caveats to its marketing.
The Mac mini and Mac Studio ran short first. The Air is the volume seller, so it is the last domino rather than the first.
Why the biggest buyer runs out
Three companies, Samsung, SK Hynix and Micron, make the overwhelming majority of the world's DRAM. Their most profitable product right now is not the DDR5 that goes in laptops and servers. It is high bandwidth memory, the stacked DRAM that sits on the package next to an AI accelerator, sold into a market that is not price sensitive and cannot get enough.
Wafer capacity is finite and fungible. Every wafer committed to high bandwidth memory is a wafer that does not become a DDR5 die, and the arithmetic of that decision is not close. IDC has estimated 2026 DRAM bit supply growth at roughly sixteen percent, which would be a comfortable year in normal times. It is not a comfortable year, because AI infrastructure demand is capable of absorbing that entire increment and asking for more.
So the queue re-forms by margin, and everyone buying ordinary memory joins the back of it. Apple's position in that queue is far better than yours. That is the part worth sitting with.
What it costs, and what it will cost
TrendForce said in July that server DRAM contract prices should rise between thirteen and eighteen percent quarter on quarter in the third quarter of 2026, with long term agreements capping the increases for the largest buyers. Read that carefully: the thirteen to eighteen percent figure is the moderated one, held down by contracts that smaller organisations do not have.
The public guidance on when this ends keeps sliding. Intel has pointed to 2028 before conditions normalise. SK Hynix has warned the imbalance could run past 2030. Neither of those is a forecast anyone enjoys giving, and neither is unreasonable given how long a new fab takes to build, equip and qualify.
For anyone writing a hardware budget, the practical consequence is that the usual instinct is inverted. Waiting for prices to fall has been sound advice in memory for most of the last thirty years. It is not sound advice this year.
What we would do
Buy memory earlier in the project than feels natural, and be willing to hold stock. If a build lands in November and the DIMMs are available in August at a known price, take the August price.
Specify to requirement rather than to comfort. The premium on larger configurations is where the shortage lands hardest, in both money and lead time, so the habit of rounding capacity up because it is only a bit more expensive has become an expensive habit.
And look at reuse before replacement, seriously rather than as a gesture. Spare DIMM slots on an existing server, a workstation that can take a second kit, a NAS whose modules can be rebalanced across nodes: all of these beat a purchase at current prices, and several of them are free. If you kept the old modules in a labelled box, this is the year that pays off.
Sources and further reading
- The global memory shortage hits the MacBook Air, TechCrunch, August 2, 2026
- MacBook Air experiencing major shortage despite 200 dollar price increase, MacRumors, August 2, 2026
- MacBook Air facing supply shortages due to the memory crisis, 9to5Mac, August 2, 2026
- Server DRAM contract prices expected to rise 13 to 18 percent in 3Q26, TrendForce, July 9, 2026
- AI boom fuels DRAM shortage and price surge, IEEE Spectrum
Frequently asked questions
Why does a laptop shortage matter to someone running servers?
Because it is the same supply, and the laptop is the visible end of it. Samsung, SK Hynix and Micron between them account for the overwhelming majority of DRAM production, and a wafer committed to high bandwidth memory for an AI accelerator is a wafer not making DDR5 for anything else. Apple is one of the largest and best resourced memory buyers on earth, with long term contracts and enormous volume leverage. When Apple starts quoting six week ship dates and adding availability warnings to its own marketing, that is a useful signal about what smaller buyers are going to face on server DIMMs, workstation kits and NAS upgrades. TrendForce is forecasting server DRAM contract prices rising thirteen to eighteen percent quarter on quarter in the third quarter of 2026.
What exactly did Bloomberg report?
Mark Gurman reported on August 2 that the M5 MacBook Air, in both the thirteen and fifteen inch sizes, is facing serious supply constraints. Delivery estimates on Apple's United States online store run from two to six weeks, and configurations with more RAM are the ones waiting longest, which points directly at memory rather than at assembly or logistics. Gurman cited retail sources describing shipments as more constrained than they can recall. He attributed it to companies building out data centers with AI servers pushing up prices for both RAM and SSD storage chips across Apple's lineup. The Mac mini and Mac Studio were affected before the Air, which fits the pattern of a shortage working its way from lower volume machines toward the highest volume one.
Is this only about DRAM, or is flash storage affected too?
Both, though DRAM is where the pressure shows first and hardest. Gurman's reporting names RAM and SSD storage chips together, and the same manufacturers sit on both sides of the business. The mechanism differs slightly: DRAM capacity is being pulled toward high bandwidth memory stacks that go on AI accelerator packages, while NAND demand is being pulled by the storage tiers those same clusters need. For planning purposes, treat any bill of materials with a large memory or flash line as exposed, and expect quotes to have shorter validity windows than you are used to.
When does this ease?
Not soon, and the public guidance keeps moving further out. Intel has pointed to 2028 before conditions normalise, and SK Hynix has warned the imbalance could persist beyond 2030. The reason is simple arithmetic rather than pessimism: new fabrication capacity takes years to build and qualify, and AI infrastructure demand has been absorbing essentially all of the annual increase in supply. IDC has put 2026 DRAM bit supply growth at roughly sixteen percent, which sounds healthy until you set it against demand that can take the entire increment. Anyone budgeting a refresh for 2027 should assume memory costs more then, not less.
What should we actually do about it?
Three things. First, buy memory earlier in the project than you normally would, and accept holding stock, because the price curve is currently working against waiting. Second, specify what you need rather than what is comfortable, since the premium on larger configurations is where the shortage bites hardest, both in money and in lead time. Third, look harder at reuse before replacement: a server with spare DIMM slots, a workstation that can take a second kit, or a NAS that only needs its existing modules rebalanced will all beat a new purchase at current prices. This is a good year to be the person who kept the old DIMMs in a labelled box.