WEBVTT

00:00:00.000 --> 00:00:03.440
If you've ever suddenly needed to buy a computer memory or an SSD,

00:00:03.440 --> 00:00:06.880
there's a good chance you've experienced either sticker shock

00:00:06.880 --> 00:00:10.560
or a pleasant surprise when you looked up how much it cost.

00:00:10.560 --> 00:00:15.440
Indeed, both RAM and the NAND flash that holds the data inside SSDs

00:00:15.440 --> 00:00:19.440
are notorious for fluctuating in price more than other components.

00:00:20.240 --> 00:00:25.280
But why? To answer, we spoke with our friends at G-Skill and we'd like to thank them for

00:00:25.280 --> 00:00:32.800
contributing their insight. As it turns out, the main reason is that both RAM and NAND flash are commodity goods.

00:00:33.440 --> 00:00:39.600
And when we say commodity, we mean a good that can be used in many different types of markets.

00:00:39.600 --> 00:00:47.280
Think about corn, for example. You can eat it, sure, but it can also be used for everything from animal feed to biofuel,

00:00:47.280 --> 00:00:52.560
so large amounts of it are regularly sold to many different types of businesses.

00:00:53.120 --> 00:00:56.320
The DRAM and NAND markets work in much the same way.

00:00:56.720 --> 00:01:02.560
Nearly all modern electronic devices, everything from TVs to smartwatches to

00:01:02.560 --> 00:01:09.840
heck, even internet-connected microwaves, need both RAM and flash storage, not just computers.

00:01:09.840 --> 00:01:14.400
By comparison, something like an x86 CPU from Intel or AMD

00:01:14.400 --> 00:01:17.840
is really only being sold to two kinds of markets,

00:01:17.840 --> 00:01:23.120
system integrators like Dell or HP and retailers like New Egg and Micro Center.

00:01:23.120 --> 00:01:28.880
You also have the fact that DRAM and NAND can be shipped off more or less unchanged

00:01:28.880 --> 00:01:33.920
to these different customers. For example, the RAM that's inside your computer

00:01:33.920 --> 00:01:37.520
is not all that dissimilar from the RAM inside your phone.

00:01:37.520 --> 00:01:44.640
So if a RAM manufacturer is trying to increase supply to serve a bunch of different use cases,

00:01:44.640 --> 00:01:51.120
all they really have to do is shrink the process node so they can fit more RAM chips on one wafer.

00:01:51.120 --> 00:01:55.520
Then they cut it up like a square-cut pizza and send however many chips to buyers

00:01:55.520 --> 00:02:01.600
serving each of these different segments. Contrast that with something like manufacturing displays,

00:02:01.600 --> 00:02:07.280
where different resolutions and panel types mean that serving many different markets

00:02:07.280 --> 00:02:11.520
is much more challenging for any single consolidated manufacturer.

00:02:11.520 --> 00:02:17.920
So this ends up meaning that more external forces are exerting influence on how much DRAM

00:02:17.920 --> 00:02:23.280
and NAND will cost you, and it also means that manufacturers have a much harder time

00:02:23.280 --> 00:02:26.560
estimating how much supply they need to keep on hand.

00:02:26.560 --> 00:02:31.120
That means that both supply gluts and shortages are more likely.

00:02:31.120 --> 00:02:34.800
Shortages can and have been caused by many different things.

00:02:34.800 --> 00:02:40.320
For example, there can be unforeseen high demand when a hot new phone comes out that everyone wants,

00:02:40.320 --> 00:02:45.600
or even a service that causes a spike in demand on the server side of things.

00:02:45.600 --> 00:02:49.120
I mean, think how much more memory Google might suddenly require

00:02:49.120 --> 00:02:53.040
if their Stadia game streaming service were to take off. Good luck with that.

00:02:53.040 --> 00:02:58.160
Other times, supply can be strained by production problems such as power outages that disable

00:02:58.160 --> 00:03:02.000
the systems used to maintain their dust-free environments on the production line,

00:03:02.000 --> 00:03:06.960
meaning that manufacturers have to scrap whatever's on the line and start from scratch.

00:03:06.960 --> 00:03:12.000
Now this might only set things back a week or two and have a minimal impact on prices,

00:03:12.000 --> 00:03:16.720
but other times, natural disasters like earthquakes and fires have shut down

00:03:16.720 --> 00:03:21.120
fabs for months at a time, resulting in significant shortages in supply,

00:03:21.120 --> 00:03:24.880
something that notably happened to SK Hynex back in 2013.

00:03:24.880 --> 00:03:29.760
There's also the possibility of politics muddying things up as they so often do.

00:03:29.760 --> 00:03:34.000
For example, there was a trade dispute between Japan and Korea in 2019.

00:03:34.560 --> 00:03:39.120
Japan supplies lots of raw materials used in the manufacture of DRAM and NAND,

00:03:39.120 --> 00:03:44.160
meaning that the fabs in Korea couldn't get the ingredients they needed for a short while.

00:03:44.720 --> 00:03:52.160
Now that situation was fortunately resolved quickly, but any kind of more protracted conflict could have driven prices up in a big way.

00:03:52.160 --> 00:03:56.720
Of course, sometimes mishaps end up working in the consumer's favor instead.

00:03:56.720 --> 00:04:02.000
In the mid-2000s, there was a huge surge of people buying mobile devices like phones,

00:04:02.000 --> 00:04:07.360
and the RAM manufacturers assumed that this trend would just keep going the way that it was.

00:04:07.360 --> 00:04:12.560
So they started churning out lots more RAM, but instead, the market got saturated.

00:04:12.560 --> 00:04:16.720
Suddenly, everybody already had a phone. Or two.

00:04:16.720 --> 00:04:20.160
And sales slowed down, so suddenly there was a lot of excess supply,

00:04:20.160 --> 00:04:23.920
meaning that folks who bought RAM during this time ended up getting a bargain.

00:04:23.920 --> 00:04:28.480
There have even been times that memory prices have been pushed below the manufacturing cost

00:04:28.480 --> 00:04:33.760
for the DRAM chips, because it's worse for the manufacturers to shut down the fabs

00:04:33.760 --> 00:04:38.640
than to keep pumping out chips at a loss like during the Great Pringles Incident of 1994,

00:04:39.360 --> 00:04:44.560
which actually wasn't a real thing. It was never a Pringles incident that I know of.
