Memory to Hit 62% of Chip Revenue in 2027, TechInsights Says

Memory will pull in roughly 62% of global semiconductor revenue next year, according to TechInsights, as AI infrastructure spending pushes the total chip market past $2.6 trillion in 2027.

The numbers are stark. TechInsights' 2027 Outlook Report, released Oct. 29, projects memory revenue above $1.6 trillion next year. That's more than double the $740 billion the segment is expected to book this year. Growth like that doesn't come from steady demand — it comes from shortage.

The driver is AI infrastructure investment. Building out data centers has tightened supply of high-bandwidth memory (HBM), server DRAM and NAND flash, and prices are climbing sharply as buyers compete for limited wafers. HBM, stacked DRAM designed to feed AI accelerators, has been the flashpoint for two years. Now the squeeze is spreading.

Where does the next bottleneck land?

NAND flash, TechInsights argues. "We expect NAND to become the next memory bottleneck as AI drives rapid growth in demand for solid-state drive (SSD) storage," the firm said in its report. AI workloads generate enormous datasets, and those datasets live on SSDs. If NAND supply can't keep pace, the price pressure that has defined HBM moves down the stack into storage.

That has practical consequences for system builders. NAND shortage means SSD prices rise. Rising SSD prices raise the cost of servers, PCs and consumer devices that rely on flash storage.

What about advanced packaging?

The report also weighs capacity at TSMC, Taiwan's dominant foundry. TechInsights projects production capacity for TSMC's advanced packaging technologies — Chip-on-Wafer-on-Substrate (CoWoS) and System-on-Integrated-Chips (SoIC) — will grow by 20% to 25% annually. Those platforms package AI processors alongside HBM stacks, so their capacity effectively sets a ceiling on how many AI chips ship.

The projection is a forecast, not a confirmed fab figure. But even at that growth rate, supply stays tight. "Even so, demand will remain strong," TechInsights said.

Who feels it first?

David MacQueen, director of insights at TechInsights, frames the shift as a procurement problem, not just a technology roadmap problem.

"Next year, companies will need to think not only about which technologies to develop but also about which memory they can secure," MacQueen said. "This will affect everything from chip design and capital investment to the consumer products that ultimately reach the market."

That's a notable reframing. For most of the past decade, memory was a commodity buyers sourced at the last minute. If TechInsights is right, memory allocation becomes a strategic input — something designers plan around at the architecture stage, the way they've long planned around leading-edge logic capacity at TSMC.

The ripple effects touch several groups at once:

• Chip designers, who may need to design for the memory they can actually buy, not the memory they'd ideally specify.
• Capex planners, whose capital investment decisions now hinge on securing memory supply, per MacQueen.
• Consumer product makers, who face component costs shaped by AI data center demand far upstream.

How solid are the numbers?

Worth separating what's measured from what's projected. The $740 billion memory revenue figure for this year is TechInsights' current-year estimate. The $1.6 trillion figure for next year, the $2.6 trillion total market size and the 62% memory share are all forecasts from the Oct. 29 report. The 20-25% CoWoS and SoIC capacity growth rate is likewise a projection of annual expansion, not a confirmed foundry capex announcement.

All of it rests on AI infrastructure investment continuing at its current pace. That's the report's central assumption — and the main risk to its numbers.

If memory does cross $1.6 trillion next year as forecast, the competitive dynamics of the whole industry tilt: Samsung, SK hynix and Micron would control the largest single slice of semiconductor revenue, and every downstream buyer — from AI chip giants to smartphone makers — would be negotiating for allocation rather than shopping on price.