Billions are flowing into memory fabs. Why chip prices won’t cool anytime soon

For decades, consumer technology followed a predictable and reassuring script: when a new model arrived, older devices became cheaper. That safety net is disappearing.

Prices of smartphones, laptops and other electronics are now rising in the middle of product cycles, long before their successors reach the market, as sharply higher component costs ripple through the technology supply chain.

At the centre of the shift is memory.

According to TrendForce, memory chip prices surged by as much as 98% in the first quarter of 2026, with a further increase of 58-63% projected in the current quarter.

The impact is already visible for consumers. Apple recently raised prices across its Mac lineup globally. In India, the entry-level MacBook Neo increased from Rs 69,900 to Rs 79,900, a jump of 14.3% that reflects the changing economics of hardware.

The memory industry is responding. Samsung Electronics, Micron Technology and SK Hynix are investing tens of billions of dollars in new manufacturing capacity. But the shortage is unlikely to end anytime soon.

AI takes priority

The rapid growth of High Bandwidth Memory, or HBM, is reshaping one of the semiconductor industry’s most cyclical markets.

HBM is an advanced type of memory used alongside artificial intelligence accelerators. As these chips process enormous datasets, they need large amounts of high-performance memory to continuously supply data to graphics processing units.

That has made HBM the memory industry’s fastest-growing and most profitable product segment.

“HBM and data centre DRAM are undersupplied due to high demand from hyperscalers and cloud service providers. All three memory vendors (Micron, Samsung and SK Hynix) are prioritising wafer starts towards HBM because of the margin differential in pricing. Thus, commodity DRAM supply is being structurally tightened by HBM conversion,” said Parv Sharma, senior analyst at Counterpoint Research.

The shift towards HBM means fewer manufacturing resources are available for conventional DRAM, which is widely used in smartphones, computers and servers. As supply has tightened, prices and profit margins for traditional DRAM have also risen.

NAND, the flash memory used to store data in devices, is receiving less incremental capital expenditure than DRAM because its recovery in profitability began more recently, Sharma said. In effect, advanced DRAM and HBM receive the bulk of new investment, while NAND gets what remains.

SK Hynix has disclosed that about 30% of its DRAM wafer starts are now dedicated to HBM, a figure expected to rise to about 40% by 2027, Sharma added.

A multi-billion-dollar expansion

To ease the structural shortage, the memory industry has embarked on one of its largest investment cycles in decades.

A global capacity race is now underway, with Samsung, SK Hynix and Micron expected to spend heavily on new factories and supporting infrastructure over the next several years.

Samsung and SK Hynix last week announced plans for joint investments exceeding $500 billion to build four mega-fabrication plants in South Korea, along with dedicated advanced-packaging hubs.

Micron, meanwhile, is pursuing one of the most ambitious manufacturing expansions in its history.

In the United States, the company has begun constructing a leading-edge DRAM factory in Boise, Idaho. It is also advancing plans for a manufacturing campus in Clay, New York, which is expected to house multiple DRAM fabs.

Outside the US, Micron is expanding advanced DRAM production at its Hiroshima facility in Japan and investing across other manufacturing locations to support growing demand from AI infrastructure.

The memory shortage has also come under legal scrutiny. A lawsuit filed in the US District Court for the Northern District of California accuses Samsung, SK Hynix and Micron of unlawfully limiting DRAM supply and driving up prices.

New fabs take years

For all the investment announcements, the industry’s biggest constraint is not capital. It is time.

Building the physical structure of a semiconductor plant is only the beginning. A modern, leading-edge memory fab typically takes three to five years to move from groundbreaking to high-volume manufacturing.

“Construction and cleanroom installation usually require 18–30 months, followed by 9–15 months for equipment installation and qualification,” said Manish Rawat, semiconductor analyst at TechInsights.

Long equipment-delivery cycles add to the delay. Extreme ultraviolet, or EUV, lithography machines made by ASML typically have lead times of 18-24 months, Rawat said.

Brownfield expansions, which add capacity at existing facilities, can be completed faster and usually take two to three years. But such projects are constrained by available floor space, electricity and other infrastructure, according to Sharma.

The bigger challenge is not simply erecting a factory. It is achieving competitive production yields and securing approval from customers.

That process can take another 12-18 months or longer.

The difficulty is particularly pronounced for advanced DRAM and HBM. More complex manufacturing processes, including EUV lithography, higher layer counts, through-silicon vias and advanced packaging, extend the time needed to reach profitable, high-volume production, Rawat said.

Through-silicon vias, or TSVs, are microscopic vertical connections that allow multiple layers of memory to be stacked and linked. They are critical to HBM performance but make manufacturing and packaging more difficult.

From glut to shortage

The obvious question is why memory makers did not begin expanding earlier.

Only three years ago, the industry was battling the opposite problem.

As the pandemic-era boom in PCs and smartphones faded, device makers were left with excess inventory. Orders dried up, while DRAM and NAND prices collapsed.

Memory manufacturers responded by cutting wafer production, postponing capital expenditure and shelving expansion plans to contain mounting losses.

As inventories returned to normal, however, the rise of generative AI created an unprecedented surge in demand for HBM and data-centre DRAM. Manufacturers that had only recently reduced output were suddenly scrambling to add capacity.

The speed of that reversal helps explain why the industry cannot quickly build its way out of the current crunch.

Relief remains distant

The investment wave will eventually add significant capacity, but analysts do not expect it to end the shortage soon.

“Current investment plans represent a substantial expansion of industry capacity, but AI demand is growing at an exceptionally rapid pace. While these investments should ease the severity of shortages over time, demand is expected to outpace supply for much of the expansion cycle. Rather than returning to prolonged oversupply, the market is likely to experience periodic supply constraints over the next five years,” Rawat said.

Billions of dollars are being poured into memory factories, but AI’s appetite for chips is expanding just as quickly.

Until that gap narrows, consumers and businesses are unlikely to see meaningful relief from rising memory costs.