Kelt Reeves has sold high-end personal computers for 34 years, catering to gamers and others who are willing to pay $4,000 and more for extreme performance.
His company, Falcon Northwest, builds these systems to order using powerful computer chips, particularly a variety known as random-access memory, or RAM, which hold data temporarily while other chips crunch numbers and display graphics.
But as OpenAI, Meta, Google and other tech giants battle to win the artificial intelligence race, they have also demanded more memory chips for the data centers they are building to power the technology. Since late summer, Reeves has been grappling with a tripling of the cost of memory chips, pushing Falcon Northwest to raise the price of some of its popular high-end computers to more than $7,000 from about $5,800.
“This isn’t a consumer-driven bubble,” Reeves, 55, said. “Nobody is expecting this to be a quick blip that’s going to be over with.” The crunch in memory chips is the latest domino effect from the AI frenzy, which has upended Silicon Valley and lifted the fortunes of AI chipmakers like Nvidia. The boom has now gone beyond AI chips to reach other components used to build gadgets, which could ultimately affect the prices of mass-market personal computers and smartphones, too.
Memory chip manufacturers can make more money selling expensive chip varieties to AI data centers than to the PC and smartphone companies that had long driven their revenue. As the chipmakers focus on producing more for AI customers, their shipments of consumer-grade chips have slowed, and their prices have surged — costs that could eventually be passed on to consumers.
TechInsights, a market research firm, predicts that higher costs for memory chips would raise the price of a typical PC by $119, or 23 per cent, by this fall from the same period last year. Besides RAM, those figures include price increases for the chips called NAND flash memory, which provide long-term storage in computers and phones.
Memory chips have long been considered a strategic commodity, an essential element in electronics that has prompted past trade wars between the United States and both Japan and South Korea.
Since memory chips are largely interchangeable, vendors for decades mainly competed on price. They often built too many factories and churned out too many chips, leading to plummeting prices and heavy losses.
Multiple makers dropped out, leaving South Korea’s Samsung Electronics and SK Hynix competing with Micron Technology, the sole US supplier. (Some Chinese RAM makers have recently entered the market.) After many boom-bust cycles, Samsung, SK Hynix and Micron are now enjoying sustained sales growth, and their share prices have soared to record highs in the past year.
The demand is “almost exclusively from AI,” said Jim Handy, whose research firm, Objective Analysis, tracks the memory market.
The memory chip boom is also boosting the overall chip industry. The International Data Corp., a research firm, projected that total chip industry revenues would jump 28%, to more than $1 trillion, this year, a milestone that many experts had not expected until 2030.
Micron, based in Boise, Idaho, is a microcosm for the changes. For 29 years, the company has sold different types of memory products to consumers and small businesses like Falcon Northwest.
But last month, Micron said it would discontinue that direct-to-consumer business, called Crucial, “to improve supply and support for our larger, strategic customers in faster-growing segments” such as AI chips.
Micron is also investing heavily in the factories that turn silicon wafers into chips. That includes building two new factories in Boise; buying an existing factory in Taiwan; and spending $100 billion on a project near Syracuse, New York, that is considered the state’s largest private investment.
The first of the new factories will not come online until mid-2027. In the meantime, Micron said, customers are receiving half to two-thirds of the memory they want to buy.
Furious demand from AI is not just for holding data in more computers. Memory is also playing a more strategic role, with new kinds of chips determining how fast applications like chatbots deliver answers.
The best example is what the industry calls high-bandwidth memory, or HBM, which is assembled by stacking RAM chips on top of one another rather than placing them side by side. Data moves vertically through layers of the stack instead of links along the edge of chips, a faster path.
Nvidia, which has been setting many hardware trends, buys huge volumes of HBM. The company bundles 12-high stacks of those chips inside the same package as its better-known AI processors, which allows faster data transfers than RAM placed elsewhere in a system.
This month, Nvidia described how a forthcoming microprocessor and an AI chip would use more and faster memory to achieve blazing speeds and lower costs per calculation.
This all means tough choices for consumers, particularly the hobbyists who assemble or upgrade their own PCs. A typical PC kit containing plug-in modules of RAM chips that sold for around $105 in early September was $250 at the end of December, according to the website PCPartPicker. — The New York Times
GRAPH POINTS
1. Memory chip manufacturers can make more money selling expensive chip varieties to AI data centers than to the PC and smartphone companies
2. Total chip industry revenues would jump 28%, to more than $1 trillion, this year, a milestone that many experts had not expected until 2030
3. Memory is also playing a more strategic role, with new kinds of chips determining how fast applications like chatbots deliver answers
4. A typical PC kit containing plug-in modules of RAM chips that sold for around $105 in early September was $250 at the end of December
5. Higher costs for memory chips would raise the price of a typical PC by $119, or 23 per cent, by this fall from the same period last year
By Don Clark
Oman Observer is now on the WhatsApp channel. Click here
