Memory-makers' shares are down. Some RAM prices have eased. Blaming Google is not a good idea
Chocolate Factory boffins have found a way to reduce AI’s memory use, but don’t assume that means less demand for DRAM
The high cost of memory has sideswiped the technology industry, causing server vendors to admit their quotes are guesstimates and depressing sales of PCs and smartphones. Nobody is immune: Microsoft used the RAM panic as cover for fixing Windows 11’s memory gluttony, and Sony suspended orders for compact flash and SD cards because it can’t buy the chips to build them.
Demand for artificial intelligence infrastructure created the situations described above by giving memory-makers incentive to production of high-bandwidth and high-margin memory GPUs require. Reduced supply for other memory sent prices soaring.
Yet over the last week, the price of consumer-grade memory has reportedly eased at some online vendors.
Memory-makers’ share prices slipped sharply. The value of Micron Technology’s scrip has slumped in the dozen days since it announced enormous growth in revenue and profits. Western Digital’s share price fell 8.5 percent on Monday alone and is down 20.5 percent since March 19th. SanDisk slid seven percent on Monday, and the company has lost a fifth of its value in a fortnight.
Those falls are steeper than those recorded by major stock market indices, as investors try to understand the impact of war in the Middle East.
- Datacenter batteries are selling years in advance, because AI, says Panasonic
- Your next car might need 300 GB of RAM, and so will autonomous robots
- Tencent says small clouds can’t get hardware, so big clouds can hike prices
- Alibaba Cloud hikes prices by up to 34%, blames hardware costs and AI demand
Given the AI boom continues largely unabated, why are investors worried?
Some have linked the change in market sentiment to a technology Google revealed last week called TurboQuant that the company’s researchers describe as “a compression algorithm that optimally addresses the challenge of memory overhead in vector quantization” – one of the most memory-hungry parts of AI workloads – and by doing so speeds up applications and “lowers memory costs.”
Google’s announcement also claims TurboQuant can reduce the amount of memory needed in key-value cache - “by a factor of at least 6x”.
Some have concluded that TurboQuant means demand for memory will fall and linked Google’s announcement of the tech to share market moves.
Analyst firm TrendForce, which specializes in the memory market, disagrees.
In a report published last week, TrendForce predicts TurboQuant will lower the cost of AI infrastructure, and by doing so “spark massive long-sequence application demand, comprehensively driving structural growth and specification upgrades for high-bandwidth, main, and flash memory across cloud and edge platforms.”
The firm thinks that TurboQuant can reduce the cost of running inferencing workloads, and suggests that this “is likely to drive substantial demand for long-context and multi-agent architectures, further accelerating the migration of AI workloads to the edge.”
Or in other words, more efficient AI will create demand for more AI, and more memory.
The war in the Persian Gulf means it will be hard to meet that demand, because the conflict has damaged the supply chain for helium – a vital component in semiconductor production that could mean chipmakers can’t make all the RAM they assumed would inflate their revenue and profits.
Share prices reflect investors’ opinions of a business’s future prospects. And helium shortages are an obvious indicator of reduced future production and sales. ®
