How the Chip Shortage Impacts Memories

A close-up of the right side of the chart below, showing a gap of 45% above trend.

There’s been a lot of talk in the press for over a year about the chip shortage, but this isn’t something that I have written about in The Memory Guy.  My coverage of the shortage has been limited to the investor posts that I write for Smartkarma, and they’re behind a paywall.  It’s about time for me to post something here.

The shortage is the reason that the world semiconductor market is constantly breaking revenue records.  Since December 2020 every month has set a new record: Each month is higher than the same month of any prior year, and often it’s the highest-ever month in the history of semiconductors.  That’s not intuitive, is it?

Here’s how it works: The shortage was caused by exceptional demand, and since this demand is not matched by production there’s a shortage, and during a shortage prices increase.  Multiply increased prices by abnormally large unit shipments and you get record revenues.

How much of a record?  Well, take a look at the chart below, which plots actual monthly revenues in red the way they’re reported by WSTS (the World Semiconductor Trade Statistics), against a 3.9% average annual growth trendline, which the industry has followed very closely since 1996:

Chart in semi-log format showing actual monthly revenues in red and a 3.9% average growth black dotted line. The red line follows the trend line most of the time, except for exceptions that are annotated in the following chart. At the far right a brace shows that February's revenues were 45% above trend.

I used a semilogarithmic format in this chart for two reasons:

        1. In a normal chart both lines would look like hockey sticks, and it would be hard to tell them apart in most years or to read the early years’ data at all
        2. Steady growth shows up as a straight line in a semilogarithmic format

The vertical axis is the log of the 3-month moving average (3MMA) of revenues in millions, so the range of this axis is $10 billion to $100 billion in monthly worldwide semiconductor revenues.

Sure, the red line is bumpy, but it generally follows the 3.9% growth line pretty closely.

What About Those Bumps?

Naturally you want to know why there are all those bumps in the chart.  I will explain.  For the most part, the small bumps are the typical ups & downs of the semiconductor cycle: They are caused by overinvestment and underinvestment in production capacity.  Demand, which matches the trend line, tends to grow very predictably.  Most semiconductor cycles are caused by overproduction and underproduction while demand steadily grows.

But in rare instances demand will undergo a big change.  Before 2008 I used to say that there was a big demand change every 15 years: 2000, 1985, and 1970.  But the 2008 meltdown shortened that cycle, and the two most recent demand changes, 2018 and 2021, have been separated by only three years.

The same chart appears below, but with the big demand changes annotated.

The same chart as above, but with three arrows pointing out the 2008-9 global financial collapse, 2018's trade war surge, and today

The 2008 Global Economic Collapse caused chip demand to vanish in the first quarter of 2009, and prices collapsed as a result.  Although demand recovered in the second quarter, prices were unable to rise back up to their prior levels, so market revenues were below those of 2008 for a year and a half.

The US-China trade war in 2017 led to big inventory growth that carried over into 2018.  This phenomenon feeds on itself: When customers buy more than they can use (in order to increase their inventory) factories can’t keep up, and that causes a shortage.  The shortage worries the customers, so they build up more inventory, only to make the shortage worse.  As the shortage worsens, customers want to increase their inventories, and so on.

Once the shortage eases, the same thing happens in reverse: Customers find that they have abnormally large inventories that they need to trim, so they cancel or postpone orders.  This motivates suppliers to reduce prices, and those reduced prices lower the value of the customer’s inventory, which, from an accounting standpoint, counts against their profits.  Naturally, this motivates the customers to further reduce their inventories, and an artificially large oversupply results.  This is what happened when prices collapsed in 2019, and semiconductor revenues fell back onto the trend line.

The current departure from the trend has been driven by the COVID-19 pandemic, which I will shortly explain in detail.

Today’s Shortage is Both Good and Bad for Memories

What makes all of this unpredictable is the interplay between memories and the chips that are in the greatest shortage, which is mainly older process nodes used for things like voltage regulators and other power-management chips.  These older technologies are so scarce largely because their business was highly predictable until 2020.  Demand growth had changed very little since the global economic collapse of 2008-9.

When I say “Older Process Nodes” I’m talking about nodes larger than 90nm (i.e. 180nm, 130nm, 0.25µm, etc.)  which are predominantly built on 8″ wafers (200mm) using a lot of fully-depreciated equipment.  This business used to grow at an extremely steady rate, but the pandemic (and probably stimulus payments) sparked a big increase in consumer electronics spending, and this caught everyone by surprise.  Capacity has had to be increased, not only in wafer fabs, but also in raw wafer production, and this requires the construction of more tools.  It typically takes tool makers over a year to construct a tool from scratch, particularly the more sophisticated tools, and it takes another several months to ramp those tools to production volume.  I’ll also mention that the tool demand has surprised the tool makers too, and their lead times have grown in response.

Bringing us back to memories, the shortage of these older technology products is limiting how many end-user systems OEMs can produce, and when they can’t produce as many systems as they could sell, then they don’t purchase as much memory as they otherwise would buy.  In brief, the shortage is constraining memory demand a little.

Something else to consider is DDR5.  This emerging interface brings a new element to the DRAM module, or DIMM—it now includes a voltage regulator chip that wasn’t on other DDR DIMMs.  Voltage regulators are in short supply these days.  This might further hamper memory growth.

Fortunately, the same COVID-driven demand that is causing the shortage of older technologies is also boosting memory demand, so the hampered memory growth is still very healthy.  It’s just not quite as strong as it would be without the shortage.

What the Future Holds

How is this likely to end?  If you have another look at the chart there seems to be every reason to expect for the market to return to its original 3.9% growth level, the black dashed line, at some point.  It always does.  That won’t be easy, since February revenues were 45% above trend.  The shortage and its record revenues will end, though, and when they do revenues are likely to follow the same steep decline that the chart makes so clear in 2019, and that also shows up to a far smaller degree in the six little downturns between 2010 and 2017.

Since memories follow commodity price swings, the memory business will be deeply impacted by the next downturn.  The timing of this downturn, though, is hard to pinpoint since the shortage is demand driven.  Supply-driven shortages are easy to predict, and that’s why Objective Analysis’ forecast is consistently the most accurate in the industry.  Demand cycles are extremely hard to call.

Of course, we’d like to bring your company this caliber of analysis, so you can put it to use in your planning.  Please contact Objective Analysis to discuss ways that we can help your company thrive in any environment.