When Apple debuts the subsequent era of iPhones this week, they’re prone to include an undesirable change: a better price ticket. A value hike from the supply-chain powerhouse could be the clearest signal but that hovering reminiscence prices have turn into unavoidable — with no finish to the reminiscence crunch in sight.
Name it “chipflation” or “RAMageddon.” The scarcity is reversing a decades-long decline in reminiscence prices that helped make client electronics extra highly effective with out making them dramatically dearer. The phrases “reminiscence costs” and “reminiscence scarcity” appeared in 473 firm transcripts final quarter, based on information offered by AlphaSense. Your entire business is seemingly engaged on the issue. And but, we’re years away from seeing it fastened.
The widespread knowledge is that AI is at fault: The expertise isn’t simply contributing to larger power costs and layoffs; it’s additionally serving to drive up the price of every part from smartphones to sport consoles.
However the scarcity was brewing nicely earlier than ChatGPT took off — and was then amplified by AI’s never-ending starvation for RAM. The present market is the product of an advanced and really worthwhile reshuffling of the reminiscence business inside a system that was already failing to maintain up.
“We have to construct extra wafer capability,” Manish Bhatia, president and COO of Micron — one of many three largest reminiscence producers — informed The Verge. “(It’s) a really totally different problem for the business than it had been for a few years earlier than, the place expertise alone was capable of sustain with the demand.”
For years, reminiscence producers might improve manufacturing by becoming extra chips onto every wafer. However these positive aspects had been shrinking and taking longer to attain. In 2021, at the same time as pandemic-era electronics demand surged, Micron concluded that the issue was extra basic: Technological advances alone would not create sufficient capability to maintain tempo with long-term demand. Producers must course of extra wafers — and construct huge new amenities to do it.
Then the reminiscence enterprise collapsed. Pandemic-era purchases of computer systems, tablets, and telephones had pulled demand ahead; client spending weakened, and producers had been left with extra stock. They misplaced cash and slowed their enlargement plans.
By the point the market started recovering, generative AI had unleashed a wave of demand far bigger and extra memory-intensive than producers had ever anticipated.
The reminiscence enterprise can be terribly concentrated. Three producers account for about 90 p.c of the market, based on Counterpoint, leaving the world depending on a handful of corporations to divide restricted capability between AI infrastructure and client units. Counterpoint estimates that Samsung managed 39 p.c of the reminiscence market within the second quarter of 2026, adopted by SK Hynix at 26 p.c and Micron at 25 p.c.
The reminiscence inside your cellphone or pc broadly falls into two buckets. DRAM, or dynamic random-access reminiscence, quickly holds the knowledge a tool wants whereas opening apps, loading webpages, or working software program. NAND flash reminiscence offers longer-term storage for issues like pictures and information.
“It’s not so simple as saying information facilities are consuming RAM. The RAM just isn’t the identical.”
Inside AI information facilities, specialised processors rely closely on a type of DRAM known as high-bandwidth reminiscence, or HBM.
“It’s not so simple as saying information facilities are consuming RAM,” David Naranjo, affiliate director at Counterpoint, informed The Verge. “The RAM just isn’t the identical.”
By stacking reminiscence chips collectively and utilizing superior connections and packaging, HBM can transfer huge portions of information to and from these processors a lot sooner and extra effectively.
It’s tougher to provide, but additionally extra profitable to promote. Deep-pocketed AI chipmakers like Nvidia and AMD and tech giants like Meta and Microsoft have an insatiable want for it to energy their AI methods — and are very happy to pay.
HBM additionally consumes considerably extra manufacturing capability than standard DRAM. Reminiscence chips are made many at a time on massive, round silicon wafers. As a result of a completed HBM product stacks a number of, bigger chips collectively, it requires significantly extra silicon. Micron estimates that producing a given quantity of HBM requires roughly thrice as many wafers as producing the identical quantity of standard DRAM.
The incentives more and more favor AI. Reasonably than guessing what number of telephones or laptops will promote a yr from now, reminiscence makers can lock in multiyear commitments from among the world’s richest corporations.
“We now have been participating in discussions with clients, prioritizing those that can assure dedicated future captive demand,” Jaejune Kim, Samsung’s government vice chairman of reminiscence, stated on the corporate’s final earnings name.
AI can be growing demand for standard DRAM. Cellphone and PC makers need to run smaller AI fashions immediately on their units, requiring extra refined standard reminiscence — and extra of it. Future fashions might have considerably extra reminiscence than their predecessors to assist these capabilities.
“The three huge reminiscence guys, they’re simply mainly allocating the capability that they’ve to those corporations.”
“As AI spreads throughout varied providers comparable to search, coding, and productiveness instruments, the scope of demand is widening from a reminiscence perspective,” stated Tune Hyun-jong, president of SK Hynix, one other of the massive three reminiscence producers, on the corporate’s most up-to-date earnings name. “We’re witnessing a structural shift in demand the place each AI reminiscence and standard reminiscence are rising collectively.”
Samsung and Micron have each stated they continue to be dedicated to traditional DRAM. Micron’s Bhatia informed The Verge that it nonetheless accounts for a lot of the firm’s wafer capability. However restricted provide, surging demand, and AI clients prepared to make long-term commitments have made it terribly profitable to cede house to HBM.
SK Hynix posted a file 76 percent operating margin final quarter, up from 41 p.c a yr earlier, whereas Micron’s adjusted gross margin hit a file 85 percent. Samsung’s semiconductor income, in the meantime, jumped roughly 250-fold from a yr earlier.
“It’s extra worthwhile,” Naranjo stated. “The three huge reminiscence guys, they’re simply mainly allocating the capability that they’ve to those corporations.”
Counterpoint estimates that DRAM costs for smartphones elevated roughly 56 p.c within the first quarter of 2026 in comparison with 1 / 4 earlier, and grew round 83 p.c within the second quarter. For 16GB of DRAM for a smartphone, Counterpoint estimates that it price roughly $42 within the second quarter of 2025 and about $181 a yr later — a rise of greater than 300 p.c.
These aren’t essentially Apple’s costs — few corporations have extra leverage with suppliers — however the figures present how dramatically the part prices of constructing a high-end cellphone have modified.
The apparent answer is to make extra reminiscence. The issue is that it takes years to perform — even when you’re working quick.
Micron is presently demonstrating simply how a lot work “extra” entails.
In July, the corporate poured the first concrete for its deliberate manufacturing advanced close to Syracuse in upstate New York. As soon as accomplished, the positioning will comprise 2.4 million sq. ft of cleanroom house, making it the most important semiconductor manufacturing web site in US historical past by that measure — and Micron’s largest facility wherever on the earth.
These cleanrooms the place the chips are made will occupy solely a fraction of your complete construct, which is roughly the dimensions of 350 soccer fields.
“By the point we’re finished, we’ll have constructed in all probability 15 to twenty million precise sq. ft of constructing house to assist the two.4 million sq. ft of cleanroom,” Bhatia informed The Verge. In one in all Micron’s fab designs, the cleanroom occupies only one ground of a five-story constructing.
Micron doesn’t count on significant output till 2030
The remaining house will home an enormous community of water-treatment, energy, air-handling, and utility-distribution methods wanted to maintain it working.
Although Micron reached the concrete pour forward of schedule, it nonetheless took greater than two years. The corporate started allowing after its CHIPS Act grant for New York was confirmed in March 2024, broke floor in January 2026, and poured concrete about six months later.
The bodily development unfolds in three broad phases. Making ready the bottom can take six to 9 months, Bhatia stated. Constructing the concrete construction and facade can require one other 18 months. However probably the most sophisticated stage comes after the constructing is standing: putting in the huge community of mechanical, electrical, plumbing, and piping methods required to function a semiconductor fab.
These methods should flip an infinite industrial constructing into an surroundings the place a single microscopic particle doesn’t land within the unsuitable place and damage a chip.
Bhatia supplied a way of the precision required:
“Think about attempting to construct all of New York state and never have any ants wherever in your complete state,” he stated. “That’s what we’re attempting to do on each one in all these wafers, and we do it again and again.”
Finishing the constructing nonetheless doesn’t produce reminiscence. Micron expects to activate and take a look at important methods like energy, water, and air flow on the finish of 2028, then spend 2029 shifting in manufacturing tools and starting pilot manufacturing. The corporate doesn’t count on significant output till 2030. Even on an accelerated timeline, the journey from allowing to manufacturing can have taken roughly six years.
The corporate’s new Idaho fab is additional alongside and is predicted to start wafer output in mid-2027.
Micron is spending huge quantities to maneuver the method alongside extra shortly, roughly doubling its capital expenditure from a yr earlier to exceed $25 billion this yr.
Even with that funding, Micron can’t determine a transparent level at which provide will catch up as a result of demand continues to develop nearly as shortly as producers can broaden.
“We’ve stated that proper now it’ll be past 2027, and we don’t see when it closes simply because demand continues to develop so quick,” Bhatia stated.
Counterpoint doesn’t count on capability to meet up with demand till late 2027 or early 2028 within the best-case situation. The market information agency IDC equally expects the scarcity to final by 2027 and nicely into early 2028 earlier than there’s significant reduction.
PC and smartphone makers can’t soak up part prices which have multiplied a number of instances over indefinitely. They’ll increase costs. They’ll make fewer units. They’ll put much less reminiscence in them. Or they will concentrate on dearer units with sufficient revenue margin to soak up the added price.
Citing hovering reminiscence prices, Microsoft not too long ago raised Xbox prices by another $100 to $150, leaving some fashions as a lot as $300 dearer than at launch. Microsoft additionally raised the value of some Floor Professional laptops and tablets by $500 over their original starting price. Meta, in the meantime, added $100 to the price of its Quest 3 headset.
Apple is maybe probably the most revealing take a look at case as a result of it ought to be among the many corporations finest outfitted to face up to the scarcity. Its scale provides it extraordinary leverage with suppliers, and its clients have traditionally proved prepared to pay premium costs. But Apple has already raised prices throughout its lineup of Macs and iPads.
Now consideration is popping to the iPhone.
Reasonably than releasing its full lineup without delay this fall, Apple is predicted to introduce the iPhone 18 Professional and Professional Max alongside its first foldable iPhone, whereas holding the usual iPhone 18 and refreshed Air till spring 2027, based on Bloomberg.
Tim Prepare dinner described reminiscence pricing as a “100-year flood”
That leaves Apple’s fall lineup concentrated closely at the costliest finish of the market — precisely the place it has probably the most room to guard its margins. Counterpoint’s supply-chain monitoring exhibits Apple making ready roughly 10 p.c extra iPhone 18 Professional and Professional Max models for September by December than it allotted to the iPhone 17 Professional and Professional Max throughout the identical interval final yr.
On Apple’s final earnings name, the corporate forecast slower development due partly to worsening provide constraints, together with for reminiscence chips, and stated larger reminiscence prices had immediately lowered its gross margin. Then-CEO Tim Prepare dinner described reminiscence pricing as a “100-year flood,” citing “exponential will increase” that had induced Apple to boost costs.
The iPhone 18 Professional might begin at $1,299, based on The Wall Street Journal’s calculation based mostly on estimates from TechInsights that consider part prices and try to attenuate margin reductions. That may be $200 greater than the iPhone 17 Professional. Samsung and Google have every raised the beginning costs of some new telephones by $100 this yr, although a number of additionally include extra storage.
Apple could also be higher positioned than most corporations to drag off larger costs. Its put in base, tightly built-in ecosystem, and constant clients make its telephones behave extra like luxurious items than interchangeable electronics. The corporate has already gained market share by maintaining its costs regular whereas lower-end rivals raised theirs. In locations like China, the world’s largest smartphone market, it nearly turned widespread sense to go for a premium iPhone when lower-cost options had been promoting for practically as a lot.
For different producers, the choices are worse.
Counterpoint and IDC are already seeing smartphone corporations cut back anticipated shipments whereas shifting towards dearer fashions. If they will’t promote as many telephones, they will at the least attempt to extract extra income from every one.
Which means extra OLED shows, AI options, bigger reminiscence configurations, and different upgrades that may justify larger costs. In consequence, income for PCs and smartphones might truly stay flat and even rise as unit gross sales decline.
In different phrases, AI isn’t simply making electronics price extra. The reminiscence scarcity is altering which electronics corporations hassle making within the first place.
Finally, extra provide ought to arrive.
Samsung, SK Hynix, and Micron are all investing closely in extra manufacturing. However as a result of new amenities take years to construct, that spending gained’t deliver instant reduction.
SK Hynix is investing 600 trillion gained in its Yongin Semiconductor Cluster, a large advanced south of Seoul first introduced in 2019, and has accelerated its goal for completion from 2045 to 2033. The corporate says it’s going to double its capability inside 5 years however nonetheless expects demand to outpace provide by 2030.
Samsung and SK Hynix additionally plan to take a position a mixed 800 trillion gained, or about $588 billion, in 4 new memory-chip factories in southwestern South Korea, based on The Wall Street Journal. Micron’s New York and Idaho tasks are a part of the greater than $250 billion it says it’s going to spend on US manufacturing and analysis by 2035.
A doubtlessly necessary fourth participant can be rising in China. CXMT is constructing extra fabs, although its expertise isn’t but thought of as superior as that of Samsung, SK Hynix, or Micron. Apple is presently testing CXMT chips however would wish approval from the Trump administration to do enterprise with the corporate.
The opposite risk is that offer gained’t must catch up as a result of demand will collapse first. If the AI funding bubble bursts, reminiscence producers might immediately discover themselves dealing with very totally different economics.
For now, that is still hypothetical. The long-term contracts AI corporations have signed with reminiscence suppliers make the instant future simpler to foretell.
“2027 is fairly locked and loaded,” Naranjo stated.
Counterpoint and IDC count on value will increase to gradual from the breathtaking tempo of the previous yr to single or low double digits. However slower will increase aren’t the identical as falling costs.
“To not say costs will come all the way down to 2025 ranges, however the value will increase will stabilize at a brand new regular,” Nabila Popal, a senior director at IDC, informed The Verge. That new regular is “at the least triple what they was once.”
New fabs nonetheless take years to come back on-line, whereas AI information facilities and more and more refined client units proceed to require extra reminiscence. And producers have little financial incentive to prioritize cheaper client electronics over clients prepared to pay handsomely and commit years upfront.
Naranjo’s forecast was blunt.
“It’s not a cheerful ending anytime quickly.”
Source link
