AI Memory Prices Just Cut PC Shipments by 21 Percent in a Quarter
Global PC shipments fell 21 percent year over year in the third quarter to 57 million units, according to Counterpoint Research, as AI-driven memory and component costs pushed retail prices up. The firm forecasts a further 27 percent drop in the fourth quarter. Memory supply is shifting toward higher-margin AI servers, and the same squeeze is already visible in phone and desktop hardware prices.
On this page
The quarter the bill came due
Global PC shipments fell 21 percent year over year in the third quarter to 57 million units, according to Counterpoint Research, as elevated memory and component costs pushed PC prices higher and buyers pushed back. Part of the decline is a comparison effect: the third quarter of 2025 was an elevated base, inflated by Windows 10 end-of-support purchases and tariff-related inventory adjustments. But the deeper cause is the one this blog has been tracking in hardware prices: memory supply is migrating to higher-margin AI servers, and the bill is landing on consumer devices. Counterpoint forecasts a further 27 percent year-over-year drop in the fourth quarter.
Every major vendor declined. Lenovo shipped 15 million units, down 23 percent; HP fell 31 percent, the steepest among the top five; Dell dropped 25 percent; Apple held up best among the big brands at minus 11 percent, helped by the MacBook Neo. Counterpoint's analysts are blunt about the shape of what comes next: commercial refresh cycles and the Windows migration are no longer enough to offset weakening demand, and 2027 is forecast to stabilize rather than recover, which the firm's associate director warns "should not be mistaken for recovery".
What the memory squeeze is doing to prices
The mechanism is allocation, not a shortage of fabrication. DRAM and NAND capacity is being redirected toward AI server contracts that pay more per gigabyte, and Counterpoint's own memory price tracker charts the resulting increases stretching into next year. The effects reach past PCs: The Verge's report ties the shipment decline directly to RAM price hikes, and analysts following the same data have tracked steep increases across smartphones and broadband devices. Vendors responded by culling affordable configurations to protect margins, which removed precisely the options most price-sensitive buyers were waiting for.
The timing matters too. The first quarter of 2026 actually grew 3.2 percent on pre-tariff and pre-price-hike buying, which means the market did not soften gradually; it pulled demand forward and then hit the wall.
What it means for local AI buyers
This is the same story as yesterday's DGX Spark pricing, told from the buyer's side. The 64GB tier exists because the 128GB tier rose from 3,999 to 6,950 dollars in a year, and the reason phones, laptops and desktop AI boxes are all repricing together is that they compete with data centers for the same memory wafers. If you have been planning a machine for local inference, the realistic guidance from this data is unglamorous: the wait-for-prices-to-fall strategy has been wrong for four consecutive quarters, memory-hungry configurations will keep repricing fastest, and the mid-size models that fit in modest unified memory are the ones whose hardware is not going to get cheaper.
The one soft note in the data cuts the other way. Counterpoint expects 2027 volumes flat against 2026, and flat means the industry eventually absorbs the price levels. The question for local AI is which machines exist at those prices: a market that ships 27 percent fewer PCs, concentrated in commercial fleets and premium configurations, is a thinner base for the cheap upgrade path that local AI has depended on.