As a tech journalist, it's easy for me to go ahead and criticize RAM and SSD makers for price hikes, and ignore the reasons as to why that might be happening.In reality, there's a ton of stuff that goes in our storage and memory—and I've personally been taking that for granted for too long.Now that I've gone and checked out Lexar's facilities in China by myself, I have a better understanding of why the increased demand from data centers has made all of our everyday tech more expensive all of a sudden.
I was invited by Lexar with a group of journalists to visit the company's various facilities in Shanghai, Suzhou, Zhongshan, and Shenzhen, China, from June 8th to June 12th, as part of the company's "Behind the Build" media tour for its 30th anniversary.Lexar had absolutely no input in the content of this article.How is an SSD made? Inside the cleanroom where memory becomes storage Before touring the Longforce Production Line in Suzhou, my group had to change into full cleanroom suits and pass through an air shower designed to blast away any dust or fiber that might contaminate the line.
That precaution alone tells you something about how sensitive this process is.Longforce is a joint venture between Longsys, Lexar's parent company, and Taiwan's PTI, and it functions as the manufacturing and packaging arm for both automotive and consumer storage products.Throughout the trip, Lexar was pretty proud to tell us that it's one of the few companies in the world that handles the entire process from when a wafer enters the factory to when an SSD, SD card, or RAM stick is shipped to consumers.
Close That process starts with slicing the wafer into dies, and mounting said dies and a controller chip, which Lexar designs in-house, onto a printed circuit board, followed by wire bonding and encapsulation.What impressed me most wasn't the assembly itself, which is all handled on Longforce's facility in Suzhou, but the testing that follows it.And to see that, the tour continued to Longsys's Zhongshan campus, near Shenzhen, where the company runs its innovation and quality labs.
Rather than pulling random samples from a batch, every single unit on this line gets evaluated for sustained write speeds, power consumption, and how it handles sector failures.From there, drives are subjected to thermal extremes, moisture exposure, physical shock, and X-ray inspection before anything ships.This facility is also where new products get validated against real-world devices rather than lab benchmarks alone—the company had a huge room full of pretty much anything users could buy that could take an SSD or an SD card, be it a console, a drone, a smartphone, or even a smart home display.
Close One of the things engineers showed us was a design that combines the controller, NAND chips, and power management into a single unified chip, shrinking a standard M.2 2280 drive down to the far smaller M.2 2230 footprint without sacrificing capacity.That technology has since shipped as the Lexar Play X, a PCIe 4.0 drive rated for read speeds up to 7,400 MB/s and write speeds up to 6,500 MB/s.Seeing this process with my own eyes made it clear that an SSD isn't just a chip in a case.
It's the product of sourcing, precision assembly, and exhaustive validation, all of which cost money well before a single unit reaches a store shelf.Why do they keep getting more expensive, then? AI data centers are eating the world's chip supply Touring a factory built for volume made the current price surge feel almost contradictory, until I looked at where that volume is actually going.The core issue isn't a shortage of factories.
It's that NAND flash and DRAM production capacity is being redirected toward AI infrastructure at a scale the consumer market simply can't compete with.TrendForce has projected client SSD contract prices to rise at least 40% quarter over quarter in early 2026, with tight supply expected to persist through the full year, and that forecast has already been exceeded in practice.The mechanics behind this are straightforward once you see them laid out.
Enterprise SSDs became the single largest NAND flash application segment in 2026, overtaking both client SSDs and smartphones.A single high-end AI GPU can require roughly 16TB of NAND storage to operate efficiently, and a full AI server rack can demand over a thousand terabytes of NAND capacity.Cloud providers buying at that scale simply outbid the consumer channel for the same wafers Manufacturers have responded by chasing the more profitable business.
AI infrastructure consumes roughly three times the wafer capacity per gigabyte compared to conventional memory, and every wafer diverted to high-bandwidth memory for AI accelerators reduces the supply available for standard DRAM and NAND.Some brands have gone further and wound down consumer SSD lines entirely to focus on enterprise contracts.Remember that the factories building your consumer SSDs are now also the same factories building the hardware for those data centers.
Standing on a production floor that was serving both markets, it was easy to see which side of that line is currently winning the allocation fight.Related Linux's creator says you should use ECC RAM—here's what that means The best kind of RAM is also the kind no one has, for some reason.Posts 1 By Arol Wright Is it getting better anytime soon? Don't expect a price drop before 2027 ...
or before the next decade Every person I spoke with in China framed this as a structural shift rather than a passing shortage, and the market data backs that up.Lexar itself wouldn't comment too much on concrete price predictions—the company's representatives stated that semiconductor pricing is cyclical and unpredictable, and that its way of staying competitive was focusing on its own internal R&D.Supply tightness is expected to persist through the rest of the year and into next year, with AI-related demand continuing to grow as a share of total NAND consumption and no incentive for foundries to redirect capacity away from high-bandwidth memory toward standard NAND.
That's a meaningfully different situation than the cyclical downturns the memory industry has weathered before, where prices eventually corrected once demand cooled.Part of what makes this cycle stickier is that the biggest buyers have already locked in their access.Cloud giants including Microsoft, Google, and AWS have secured long-term NAND contracts, in some cases already negotiating supply allocations for 2027, which leaves less room for spot-market relief even if demand elsewhere softens.
Overall NAND demand is projected to grow 20 to 22 percent year over year while supply grows only 15 to 17 percent, a gap that isn't expected to close in the near term.There is one wildcard worth watching.Chinese domestic NAND producers, particularly YMTC, are expanding their own manufacturing capacity, which could eventually loosen the market from a different direction than the established Korean and American suppliers.
But new fab capacity takes years to come online at a meaningful scale, and nothing I saw or heard in Suzhou or Zhongshan suggested that relief is coming within the next few quarters.If you need a drive now, buying sooner rather than later is the more rational move.Waiting for 2024-era pricing to return isn't a strategy anyone in this industry is currently betting on.
The real cost of an SSD The way an SSD is made is heavily documented online, but it's one thing to read about it and a different thing to see it by yourself.This trip was an eye-opener in many ways
Read More
