Memory Crunch Spreading
AI demand is pushing memory and storage costs higher: reports say SSD prices have spiked and industry leaders warn memory will be a dominant constraint for AI infrastructure. (theverge.com) At the same time China’s cloud players are vertically integrating — Alibaba opened a data centre running on 10,000 of its own AI chips as nations respond to compute supply pressures. (cnbc.com)
A solid-state drive is the box where a computer keeps its files when the power is off, and this week that box got a lot more expensive. The Verge reported a Western Digital Black 2 terabyte drive that cost $173 in 2024 now sells for $649, while a 4 terabyte Samsung 990 Pro that was about $320 is now near $1,000. (theverge.com) The jump is tied to artificial intelligence servers buying the same basic kind of flash storage in much larger volumes. TrendForce said on April 8 that supplier inventories of NAND flash, the chip inside most solid-state drives, had fallen to multi-quarter lows as strong artificial intelligence server demand pushed up second-quarter 2026 contract prices. (trendforce.com) Memory is different from the processor that does the math. A processor is the engine, but memory is the workbench, and TrendForce said suppliers are shifting factory capacity toward server memory and away from cheaper consumer products, with conventional dynamic random-access memory prices expected to rise 58 to 63 percent quarter over quarter and NAND flash prices 70 to 75 percent in the second quarter of 2026. (trendforce.com) That squeeze is now showing up in what executives say out loud. Bloomberg reported on March 24 that OpenAI chief operating officer Brad Lightcap said, “Right now, it’s memory,” when asked about bottlenecks for expanding artificial intelligence infrastructure in the United States. (bloomberg.com) China is answering the same bottleneck by trying to own more of the stack itself. CNBC reported on April 8 that Alibaba and China Telecom launched a data center in Shaoguan, Guangdong, built around 10,000 of Alibaba’s own Zhenwu artificial intelligence chips for training and inferencing. (cnbc.com) That matters because Alibaba is not just renting out cloud servers anymore. CNBC said the company now designs chips through its T-Head unit, builds data centers, develops its own artificial intelligence models, and sells them through Alibaba Cloud, which makes the new facility a full in-house pipeline from silicon to software. (cnbc.com) The political backdrop is just as important as the hardware list. CNBC said Chinese firms accelerated domestic chip development after the United States restricted China’s access to advanced semiconductor technology, including artificial intelligence chips from Nvidia. (cnbc.com) Alibaba and China Telecom said the Shaoguan site is expected to grow from 10,000 chips to 100,000 chips. When storage prices jump at the same time cloud companies start building larger captive systems with their own chips, the shortage stops looking like a temporary shopping panic and starts looking like the new cost of building artificial intelligence at scale. (cnbc.com)
