Artificial intelligence is an important cause, but the shortage did not begin with ChatGPT. Memory manufacturers were already struggling to expand production through technical improvements alone. For years, they increased output by fitting more chips onto each silicon wafer. Those gains became smaller and took longer to achieve.
Micron concluded in 2021 that manufacturers would eventually need more wafer-processing capacity, not merely cleverer chip designs. Then the pandemic electronics boom collapsed. Consumers had bought computers, tablets and phones early, demand weakened, inventories piled up, and memory companies lost money. Expansion plans slowed at exactly the wrong moment.
When the market recovered, generative AI brought demand far beyond earlier forecasts. The shortage was already forming. AI simply arrived with a very large order.
How AI data centres consume chip capacity
Phones and AI servers do not usually use the same type of memory. Consumer devices rely mainly on conventional DRAM for temporary working data and NAND flash for long-term storage. AI accelerators depend heavily on high-bandwidth memory, known as HBM.
HBM stacks multiple memory chips and connects them through advanced packaging, allowing enormous quantities of data to move quickly and efficiently. It is difficult to manufacture, consumes more silicon and sells at attractive margins to customers including Nvidia, AMD, Meta and Microsoft.
David Naranjo, associate director at Counterpoint, cautioned that it is “not as simple as saying data centers are consuming RAM.” The memory products differ, but they compete for overlapping factory resources.
Micron estimates that producing a given quantity of HBM requires roughly three times as many wafers as making the same amount of conventional DRAM. Its regulatory filings also warn that HBM needs more wafer and cleanroom capacity per bit, which can constrain supply for downstream markets.
AI is increasing demand for ordinary DRAM too. Phone and PC makers want to run smaller AI models directly on devices, which requires more advanced memory and larger configurations. SK Hynix president Song Hyun-jong has described this as a structural change in which demand for AI memory and conventional products rises together.
Three companies control most of the market
The memory industry offers buyers few alternative suppliers. Samsung, SK Hynix and Micron account for about 90% of the market, according to Counterpoint Research. In the second quarter of 2026, Samsung held an estimated 39% share, followed by SK Hynix with 26% and Micron with 25%.
Those companies must decide how to divide limited capacity between consumer electronics and AI infrastructure. The financial incentives are not subtle. AI customers can sign multiyear agreements, commit to large volumes and pay premium prices. Consumer-device manufacturers must forecast how many phones or laptops people might buy next year, which is less reassuring to a factory operator planning billions of dollars in investment.
Samsung executive vice president Jaejune Kim said the company was prioritising customers able to guarantee future demand. Naranjo put the calculation more directly: “It’s more profitable.”
Recent results support that assessment. SK Hynix posted a record operating margin of 76% last quarter, compared with 41% a year earlier. Micron’s adjusted gross margin reached a record 85%, while Samsung’s semiconductor profit increased roughly 250-fold from the previous year.
The companies say they remain committed to conventional memory. They also have shareholders, which tends to keep the more profitable product near the front of the queue.
What the shortage could add to an iPhone
Counterpoint estimates that smartphone DRAM prices rose about 56% in the first quarter of 2026 from the previous quarter, followed by another increase of roughly 83% in the second quarter. Its estimate for 16GB of smartphone DRAM climbed from about $42 in the second quarter of 2025 to approximately $181 one year later.
Those figures are not necessarily what Apple pays. Few electronics companies have comparable bargaining power or purchasing volume. Still, they illustrate how dramatically the economics of a high-end phone have changed.
TrendForce calculates that memory represented about 10% of the iPhone 18 Pro’s component cost a year ago but could now account for roughly 34%. Its forecast says Apple may sacrifice part of its margin to protect sales, although some retail increase is effectively unavoidable.
The Wall Street Journal, using TechInsights component estimates, calculated that the iPhone 18 Pro could start at $1,299. That would be $200 above the iPhone 17 Pro. The figure remains an estimate until Apple announces pricing.
Samsung and Google have already increased starting prices for some new phones by $100 during 2026, although several models include more storage. Globally, prices for existing smartphones have risen about 15% this year, Counterpoint says. New launches are around 25% more expensive than comparable releases a year earlier.
Apple may turn the shortage into an advantage
Apple is expected to introduce the iPhone 18 Pro, iPhone 18 Pro Max and its first foldable iPhone this autumn, according to Bloomberg. The standard iPhone 18 and an updated iPhone Air are expected in spring 2027 instead.
That schedule concentrates the autumn range at the expensive end of the market, where Apple has more room to absorb higher component costs or pass them to customers. Counterpoint’s supply-chain tracking indicates that Apple is preparing about 10% more Pro and Pro Max units for September through December than it allocated to the equivalent iPhone 17 models last year.
Outgoing Apple chief executive Tim Cook called current conditions a “100-year flood on memory pricing.” He said memory costs more than accounted for Apple’s sequential gross-margin decline and warned of further pressure in the September quarter. Apple has already raised prices for Macs and iPads.
Still, Counterpoint believes the company could tolerate a hardware-margin hit for as long as two years. Its services business, large installed base, supplier contracts and tightly integrated ecosystem provide options unavailable to lower-margin competitors.
That could make the shortage a competitive advantage. If cheaper Android phones approach premium prices, some buyers may decide an iPhone offers better long-term value. Apple has previously gained market share by holding prices steady while rivals increased theirs. Whether it wishes to fund that strategy again is another matter.
Higher prices are shrinking the smartphone market
Manufacturers cannot absorb component costs that have multiplied several times over indefinitely. Their available responses are limited:
- Raise retail prices
- Reduce production and shipment targets
- Install less memory
- Concentrate on premium models with higher margins
IDC now forecasts worldwide smartphone shipments will fall 16.7% in 2026 to just over 1 billion units. It expects shipments in the second half of the year to drop 27.2%. Counterpoint predicts that average shipment prices in North America will rise 13.5%.
Companies are already shifting their product mixes towards OLED screens, larger memory configurations, AI features and other upgrades that can support higher prices. As a result, industry revenue could remain flat or even increase while fewer phones reach customers. Selling less for more is not usually presented as a consumer benefit, for understandable reasons.
The squeeze may also encourage people to keep their devices longer or buy refurbished phones. Both responses reduce demand for new entry-level models, making the lower end of the market even harder for manufacturers to serve profitably.
The same pressure has reached other electronics. Microsoft increased Xbox prices by another $100 to $150, leaving some models as much as $300 above their launch prices. Some Surface Pro computers rose by $500 from their original starting prices, while Meta added $100 to the Quest 3 headset.
Why new memory factories cannot help quickly
The obvious solution is to manufacture more chips. Unfortunately, semiconductor factories take years to permit, construct, equip and test.
Micron poured the first concrete in July for a manufacturing complex near Syracuse, New York. Once complete, it is planned to contain 2.4 million square feet of cleanroom space, the largest semiconductor manufacturing site in United States history by that measure. Supporting those cleanrooms could require 15 million to 20 million square feet of total building space.
The project moved from confirmed CHIPS Act funding in March 2024 to groundbreaking in January 2026 and its first concrete pour six months later. Micron expects to activate and test power, water and ventilation systems near the end of 2028. Equipment installation and pilot production should follow in 2029, with meaningful output not expected until 2030.
Its expanded Idaho facility is further ahead but is not expected to begin wafer output until mid-2027. Micron plans more than $25 billion in capital expenditure this year, roughly double the previous year’s level.
Company president and chief operating officer Manish Bhatia said the shortage should continue beyond 2027, with no clear point at which supply catches demand. Counterpoint’s best-case forecast puts balance in late 2027 or early 2028. IDC also expects meaningful relief no earlier than 2028.
Massive investment still may not end the crunch
Samsung, SK Hynix and Micron are committing extraordinary sums to expansion. SK Hynix plans to invest 600 trillion won in its Yongin Semiconductor Cluster south of Seoul and has brought its completion target forward from 2045 to 2033. It aims to double capacity within five years but still expects demand to outpace supply through 2030.
Samsung and SK Hynix also plan a combined investment of 800 trillion won, around $588 billion, in four memory factories in southwestern South Korea. Micron says its New York and Idaho projects form part of more than $250 billion in planned United States manufacturing and research spending through 2035.
China’s CXMT could eventually become a significant fourth supplier. It is building additional fabrication plants, although its technology is not yet considered as advanced as that of the three market leaders. Apple is testing CXMT chips but would need approval from the Trump administration before doing business with the company.
A collapse in AI investment could ease the shortage sooner by reducing demand. For now, long-term supply contracts make that scenario less relevant to immediate pricing. Naranjo said 2027 demand is already “pretty locked and loaded.”
Price increases may slow to single or low double digits, but that does not mean prices will return to previous levels. IDC senior director Nabila Popal expects a new normal at “at least triple what they used to be.” Apple’s September event will provide the clearest evidence yet of how much of that new normal consumers are expected to finance.
