Download this article in PDF format.
The memory chip shortage is hitting industrial supply chains as artificial intelligence (AI) data centers consume more computer memory (DRAM) and flash memory (NAND) capacity. Hyperscalers like Microsoft, Google, Amazon and Meta are buying aggressively for AI servers, high-performance computing and storage-heavy workloads, leaving manufacturers, electronics suppliers and component buyers competing for a smaller pool of parts.
That puts memory chips directly in the sourcing lane for industrial controls, automation systems, embedded devices, communications equipment and connected products. If recent reports are accurate, the problem isn’t going to ease anytime soon.
“There's no quick fix, as the Great AI Infrastructure Buildout pits tech hyperscalers (think Meta, Google, and Amazon) against consumer electronics makers (think Apple, Xbox, and scores of other gadget makers) jockeying for capacity from the memory manufacturers,” Steven Tweedie writes in “The memory shortage is expected to get even worse.”
“Memory suppliers are faced with unprecedented demand and deciding which type of memory chips their manufacturing lines will make,” he adds, “and many have been seizing the moment and re-jiggering allocation in favor of the deep-pocketed AI hyperscalers.”
Getting to the Root of the Problem
In “AI’s Hidden Cost: The Global Memory Shortage Threat To Affordable Tech,” Tim Bajarin reflects back to 1985, when an IBM PC cost about $6,000 (nearly $20,000 in current terms). Today, consumers can buy a Tecno Spark Go smartphone, which is thousands of times more powerful than that legacy PC, for $30-120.
“That price-performance trajectory is unlike anything else in the history of consumer goods,” Bajarin says. “But that era is ending because memory is becoming the bottleneck. Something fundamental has changed, and the industry needs to wake up to it: computer memory (DRAM) is becoming the bottleneck that will reshape consumer technology.”
Some of the bottleneck can be traced to the amount of time and money it takes to build a new chip plant. They’re massive undertakings that can’t be spun up overnight, and they require a lot of planning, resources, money and infrastructure. Building a world-class DRAM fabrication facility costs between $15 billion and $20 billion, Bajarin points out.
Industry contraction hasn’t helped at all. The industry has been whittled down to just three major players that supply over 90% of the world’s memory chips: Samsung and SK Hynix in South Korea and Micron in the U.S.
Throw in high demand for chips from the automotive sector and consumers’ insatiable appetite for the latest-and-greatest devices, and you get a recipe for tight supply, rising prices and longer lead times.
4 Ways to Manage the Strain
With no major easing expected in the second half of the year, memory buyers should be planning earlier, talking to suppliers more often and giving engineering teams more approved options before allocations get even tighter. With AI data centers, automotive and consumer electronics sectors all competing for DRAM and NAND capacity, waiting for the market to settle down isn’t a good strategy.
Z2Data offers these four tips for buyers working through the shortage:
- Use market data to guide inventory decisions. Factor in demand forecasts, supplier reliability, current lead times and market conditions to decide which memory parts need more coverage.
- Get closer to your key suppliers. Longer-term agreements, earlier conversations and better demand signals can help buyers compete for supply when capacity gets tight.
- Broaden your sourcing options. Review approved vendor lists, qualify alternate parts where possible, and look for supplier or regional exposure that could create problems later.
- Explore alternative memory technologies. Z2Data points to MRAM, ReRAM and neuromorphic memory architectures as longer-term options that may give design teams more choices as the market changes.
