AI infrastructure spending has tipped the memory market's center of gravity toward data centers: more than half of this year's DRAM shipments are going into servers and AI systems rather than phones and PCs, according to new data from Counterpoint Research.

The shift is worth pausing on, because it inverts the logic that built the memory business. For decades DRAM flowed mostly into consumer devices; now the AI buildout is the main customer, and — as a closer look at the numbers shows — it is paying an even larger share of the bills than its volume alone would suggest.

What the Tracker Shows

In its Memory Tracker published June 29, Counterpoint said server DRAM accounted for 48% of total DRAM shipments this year and high-bandwidth memory (HBM) for 9% — together 57%. By comparison, mobile took about 22% of shipments and PCs roughly 10%. HBM, which stacks multiple DRAM chips vertically to speed data processing, is a key building block for AI data centers.

On a revenue basis, server DRAM and HBM made up an even larger 65% of the market, since data-center memory commands higher prices and added value than memory for mobile, PCs, and home appliances.

Why the Money Tilts Harder Than the Volume

That gap between the two numbers — 57% of shipments but 65% of revenue — is the part worth understanding, because it says something the headline share does not. Not all DRAM sells for the same price. The memory destined for AI servers, and especially HBM, commands far higher prices and margins per unit than the standard DRAM inside a laptop or a home appliance. So when the market is weighted by dollars rather than by bits, the data-center share climbs from 57% to 65%: the AI side is buying a little more than half the volume but paying for closer to two-thirds of the value. Counterpoint said that price-and-value premium is exactly why the revenue share outruns the shipment share.

The Other End of the Shift

That premium is also why the shift is felt well outside the data center. Because high-value server and HBM products earn so much more per wafer, memory makers have steered capacity toward them — and every wafer committed to data-center memory is one not producing the cheaper DRAM that goes into consumer devices. The result has shown up in prices: device makers have begun passing higher memory costs on to buyers, a sign of how thoroughly the AI buildout now sets the terms for the whole memory market.

Counterpoint's Outlook

Counterpoint said that gap explains why the revenue share outruns the shipment share, and that as long as AI infrastructure investment continues, data-center DRAM demand should hold a solid trajectory over the medium to long term. That outlook is a projection rather than a certainty, and it rests on the big-tech AI spending that has driven the shift staying on its current multi-year course. But for now, the data describes a memory industry whose biggest customer is no longer the device in your pocket.

Frequently Asked Questions

What share of DRAM goes to data centers?

According to Counterpoint Research's Memory Tracker published June 29, 2026, server DRAM accounts for 48% of total DRAM shipments this year and high-bandwidth memory (HBM) for another 9%, meaning 57% of all DRAM shipments are going to data centers and AI systems. By comparison, mobile devices take roughly 22% and PCs about 10%. On a revenue basis, the data-center share is even higher at 65%, because server and HBM memory sells at higher prices than memory for consumer devices.

What is HBM memory?

HBM, or high-bandwidth memory, is a specialized type of DRAM that stacks multiple memory chips vertically and connects them with dense internal wiring, allowing data to move far faster than with conventional memory. That speed is essential for AI accelerators, which must be fed enormous amounts of data quickly to keep their processors busy. Because HBM is more complex to manufacture and central to AI systems, it commands significantly higher prices and margins than standard DRAM, and it has become a key building block of AI data centers.

Why is DRAM revenue share higher than shipment share?

The two figures measure different things. Shipment share counts units (or bits) of memory, while revenue share counts dollars. Data-center memory — server DRAM and especially HBM — sells at considerably higher prices per unit than the DRAM used in phones, PCs, and appliances. So even though data centers take 57% of DRAM shipments, they account for 65% of revenue, because each unit they buy is worth more. Counterpoint attributes the gap to the higher price and added value of data-center memory products.

Why are memory prices rising?

Memory makers have been steering production capacity toward high-value data-center products, particularly HBM, which earns far more revenue per wafer than conventional consumer DRAM. Because these products share manufacturing capacity, every wafer devoted to data-center memory reduces the supply available for the cheaper DRAM used in phones, laptops, and other devices. That tightening supply has pushed up prices across the market, leading some device makers to raise prices on consumer products to offset higher memory costs.