The memory business is currently faring pretty well. DRAM spot prices have been relatively stable for more than a year, and NAND flash spot prices, although down 20% from their spring peak, are still 2½ times their mid-2023 trough.
The business is definitely off-trend, though, and that means that it’s highly unlikely to enjoy its current strength for an extended period.
So the big question today is: “How long will this uptick last?”
The current surge is a demand-driven cycle, and there’s no hiding the fact that a lot that demand is coming from massive AI purchases in hyperscale datacenters. It’s difficult for The Memory Guy to forecast the duration of such cycles. Although capacity-driven oversupplies are straightforward to forecast, demand-driven cycles tend to be caused by factors that are very hard to predict.
I’m in good company — all the career economists who work for the world’s governments and financial institutions have an equally hard time, and they’re trained in the field, whereas I was trained as an engineer.
Twenty years ago demand-driven cycles were infrequent. I used to say that they impacted the semiconductor market every 15 years: The early 1970s oil embargo, the 1985 downturn, and the 2000 Internet Bubble Burst. Then the pace accelerated, with the 2008 Global Financial Collapse, followed by demand downturns due to the US/China Trade War in 2018, and the post-pandemic Return to Work in 2022. Each of these caused unpredictable market cycles.
So now we have today’s AI-driven cycle. How long will it last? Well in August at FMS I presented a chart showing the history of hyperscaler capital expenditures (CapEx) shown below. It’s unusually high at the moment, but it’s not clear how long these companies will continue their high expenditures. They haven’t had a matching revenue surge, so they can’t fund accelerated spending forever.
Let’s assume that this cycle follows the trend of the last two demand-driven cycles. How would it look?
The figure below is based on memory revenues, and superimposes the two past cycles, 2017 and 2021, with today’s market. These cycles have all been normalized to the market’s underlying trend, so they are all expressed in a percentage relating to how far off-trend they are, rather than in absolute revenues.
The past two cycles, the 2017 cycle in red, and the 2021 cycle in black, peaked around Month 20, and then experienced a collapse. The green cycle is the current market, and the dashed part is a projection of where it might head if it performs as did the prior two cycles. Today we appear to be in Month 18.
This isn’t as scientific of a forecast as we usually produce, but it certainly provides something worth consideration. Today’s heavy AI spending can’t last forever, and when it does end, there will undoubtedly be an oversupply with a subsequent price correction, if not a collapse like those seen in 2018 and 2022. It’s a good thing to keep a watch out for.
This is the kind of analysis that Objective Analysis provides for its clients on a regular basis. Your company would benefit from such an advance warning if it is not already an Objective Analysis client. Please contact us if you would like to benefit from this valuable insight to ensure your organization’s highest level of performance.