13h ago

Carlos Perez argues agentic AI will break historical memory cycles, challenging JPMorgan's forecast of a 2026 price deceleration

JPMorgan expects DRAM and NAND price growth to slow.

0
Original post

JP Morgan Research: DRAM and NAND average selling price changes start hitting the brakes around late 2026 to early 2027. AI has hugely changed the demand story in 2025-26 for high-bandwidth memory and advanced servers, but it has not changed the basic physics of the memory business. Memory prices rose sharply when supply is tight, inventories are lean, and buyers are forced to secure capacity before someone else does. That pressure can look permanent near the top of the cycle because every customer sounds urgent and every supplier seems disciplined. Then the delayed response arrives: new capacity, technology migrations, customer digestion, and a slower rate of price increases even before absolute prices start falling. Do not confuse a stronger cycle with the end of cyclicality.

3:51 AM · May 30, 2026 View on X

JPMorgan seems tone-deaf in reading the market. We are an entirely new regime in which agentic AI means token consumption isn't based on the number of human users.

9:00 PM · May 30, 2026 · 929 Views

Memory is up not because of a cyclic trend but rather a regime change. NVidia's chips are nothing without enough high bandwidth memory. The faster the chip, the more memory it needs!

Carlos E. PerezCarlos E. Perez@IntuitMachine

JPMorgan seems tone-deaf in reading the market. We are an entirely new regime in which agentic AI means token consumption isn't based on the number of human users.

9:00 PM · May 30, 2026 · 929 Views
10:43 PM · May 30, 2026 · 242 Views