One of the biggest markets fueling today’s memory growth is hyperscale computing. Not just hyperscale computing the way that it’s always been, but hyperscale computing fueled by AI spending.
The Memory Guy decided that this might be a good time to investigate whether a market anomaly might be driving this growth — something that could change at some point to cause a memory downturn. This post was compiled to show those companies who are not already Objective Analysis subscribers the story that we have been telling our clients for the past two months.
What’s Happening in the Hyperscale Community?
Is hyperscale spending abnormally high? Yes!
Is it likely to undergo a correction? Yes, again.
Let’s look at some numbers.
There are relatively few hyperscale computing companies in the world. so it’s relatively easy to track their behavior. The chart below is similar to one I shared in a relatively recent post: How Long Will Memory Growth Continue? (Look for the most colorful graph in the post.) This is the sum of the capital expenditures (CapEx) for those leading hyperscale companies that report their financials.
The dashed black line is simply a trend to help you see when CapEx is above or below trend. Not only has CapEx recovered since the last downturn in 2022-2023, it has gone significantly above trend in the most recently-reported quarter.
This could be justified if revenues grew at a similar rate. Many companies make sure that they remain safe by setting their CapEx to be a fixed percentage of their revenues. The hyperscalers haven’t been doing this. They have been spending an increasing share of their revenues, probably under the assumption that they will recoup their investment once their customers start to actually pay for all the AI resources that they currently appear to want.
This can be charted too. If we take all of the CapEx for all of the companies shown in the graph above, and divide it by all of their combined revenue, we get the chart below:
This chart shows that these companies kept their spending around 13% of their revenues in 2021 and early 2022. When revenues dropped in late 2022, they over-responded somewhat and took spending down as low as 10.4%, but once the industry settled down, they brought their spending back up again.
Then, with the sudden popularity burst of AI in late 2023, these hyperscalers invested heavily, bringing their CapEx to a 15.1% rate that they probably can’t sustain.
Should the public’s interest in AI fall off, you can be assured that this spending level will fall, and there may even be an over-correction. At this point we can expect to see a memory oversupply, with a price fall that I will explain shortly.
Political Elements at Work
Now let’s add in another component that has impacted demand, but that is not reflected in these numbers. Since the spring, the US government has been hinting that it might try to prevent companies in China from purchasing key AI-related technologies like HBM DRAM. Major China purchasers have reacted as they did in 2017, when we had that enormous DRAM shortage that was driven by similar fears. These companies in China are building inventory to assure that they can continue to operate in case their supply is suddenly cut off. This just makes good sense.
That additional demand appears to the suppliers to be “sell-through” (parts that will immediately be used in production), and purchasers aren’t motivated to tell their suppliers when they are building inventory, since that knowledge might put the suppliers in a stronger negotiating position.
What usually happens, though, is that inventory builds cause shortages to worsen, and a worsening shortage will lead purchasers to further increase their inventory. The situation feeds on itself.
Going from a Shortage to an Oversupply
At some point supply catches up with this false demand, prices soften, and the purchasers have to take an inventory devaluation (a “Write-Down”) on their balance sheets. The value of this write-down comes straight out of the company’s profit, making the company look less appealing to stockholders, and that drives the price of the stock down. That’s the last thing a company president wants to happen.
Naturally, the company president will worry about further write-downs of a bloated inventory and will tell the company’s buyers to stop purchasing that product altogether until the inventory reaches a minimal level. This is called an “Inventory Digestion” phase, and it suddenly worsens the oversupply , which means that suppliers will be left with excess product that they can’t sell.
Their response is to lower the price to try to sell off the excess product, causing a downward price spiral, until the price reaches cost. The companies stop there, because they would get into legal trouble if they were to sell below cost. This is what happens during every memory market downturn, and it will happen again at the end of the current cycle.
Objective Analysis tracks this kind of information to give our clients insight and understanding about what will happen next, so that they can plan ahead and overtake their peers. If your company would like to gain this kind of advantage, please contact us to learn how we can work together to your benefit.