If consumers aren’t going to or are much less likely to upgrade, then that affects demand from them, and one would expect manufacturers to follow what consumers demand.
Prices rarely, if ever, go down in a meaningful degree. Stuff like this is partially necessity and partially a REALLY good excuse to see what the price ceiling actually is… and then turn that into the floor moving forward. Just look at gas prices
The “AI Bubble” is likely to be on the same level as the Dotcom Bubble and the like. It is going to be brutal and a LOT of people are going to lose their jobs… and then much of the same tech will still dominate just with more realistic expectations. And that will still need large amounts of memory
If the “AI Bubble” really is as bad as people seem to want it to be: A LOT of the vendors who make the parts you are buying RAM to use are going to be gutted. And then RAM production will drop drastically. Which will decrease supply and…
The whole injecting AI into everything has been quite annoying really. There are still many workplaces that prefer deterministic results like accounting and finance and Xero injecting AI into reconciliation or improperly trained individuals ruining the whole datasets on excel has been an absolute nightmare.
Which is why people who actually look at trends tend to compare it more to the Dot-com bubble.
The short version? A few early internet adopting sites (like Amazon…) set up online retail presences. People were ecstatic because you could now do most of the monthly shopping online and even re-buy pants that you know will fit and so forth.
Seeing money, EVERYBODY made an online retailer or service website and EVERYONE wanted to invest in that.
Then the market was oversaturated and companies with no right to exist went bankrupt and it was a bloodbath.
Except… not really. Because while the massively overinflated stock market did indeed “downturn” and a LOT of those scam companies went away, the actual fundamental premise of online first companies was a very sound one. I mean… just look at “Cyber Monday” and so forth.
And “AI” will almost definitely go the same route. Because, yeah, LLMs are HORRIBLE for accounting and finance. But they are actually really good for replacing the early career folk who translate earnings into reports. And ML in general is excellent at detecting patterns which can mean potentially billions of dollars in investing. But, like all things, it is about verification and caution. You actually need a human to read that earnings report before you send it to the investors. And you only give your “AI” a small portion of your portfolio. Same as with any team.
Prices rarely, if ever, go down in a meaningful degree.
In 2011, there was a large flood in Thailand that impacted ~40% of hard drive manufacturing. As a result, hard drives significantly increased in price. This was back when SSDs weren’t mainstream yet.
A year or two later, when manufacturing capacity was restored, prices were essentially back to what they were before the disruption.
Apart from disruptions like that, HDDs, SSDs, and RAM have always been going down in price.
What you are describing is something different… that is “close enough” to Moore’s Law for all but the most pedantic.
The (I forget the proper economics term so) base price of RAM/Storage does indeed go down as new processes and economies of scale are developed. But the cost of a “laptop hard drive” remains pretty steady in the sense that a couple hundred MB was enough back in the day but you REALLY want at least 500 gigs now. The price per byte does indeed drop rapidly but the price per “drive” is far more stable (not fully stable due to inflation and how many people are buying them, but within spitting distance).
Its why a good rule of thumb was to always just spend roughly the same on storage during an upgrade and that would result in faster technologies and larger capacity drives and so forth.
That isn’t what is happening with RAM in 2025. A much better comparison is GPUs because… it is the same problem. It is ridiculously high demand from businesses (often startups pouring dump trucks of VC money into their only hope… well, VC money or drug money in the case of miners but they matter a lot less these days) driving this. A quick search didn’t yield an easy graph and I can’t be bothered to go dig through Gamers Nexus’s twelve videos on it, but the price of an “entry level” GPU has drastically changed in the past decade.
But just for two-ish data points?
The GTX 980 and 970 had an MSRP (probably) of 550 and 330 USD, respectively, back in 2014
While there is some other bullshit involved, the RTX 5080 and 5070 have MSRPs of 1000 USD and 550 USD in 2025
Adjusting for inflation, the 980 and 970 would still only be about 753 and 451 USD in 2025 dollars
And let’s not forget that basically no cards were sold at MSRP back in early 2025…
The last point being what is, by all accounts, going to be the new normal. Barring outside impacts like… RAM going through the roof. Vendors will sell the cards for the ACTUAL MSRP rather than the inflated demand prices. And they will still be considerably more expensive as a result.
All of which is to say… my current card is definitely good enough but having a hard time deciding if I do one “final” upgrade for the decade. But I am an AMD boi so those are at least “reasonable” in terms of price per performance.
In 1995 I bought 4 MB of RAM for DM 200, which, adjusted for inflation probably works out to about €200 in today’s money. I’d say, prices have come down quite a bit, since.
If consumers aren’t going to or are much less likely to upgrade, then that affects demand from them, and one would expect manufacturers to follow what consumers demand.
RAM prices will come down sometime after the AI bubble bursts and they start making more DDR5 again.
The whole injecting AI into everything has been quite annoying really. There are still many workplaces that prefer deterministic results like accounting and finance and Xero injecting AI into reconciliation or improperly trained individuals ruining the whole datasets on excel has been an absolute nightmare.
Which is why people who actually look at trends tend to compare it more to the Dot-com bubble.
The short version? A few early internet adopting sites (like Amazon…) set up online retail presences. People were ecstatic because you could now do most of the monthly shopping online and even re-buy pants that you know will fit and so forth.
Seeing money, EVERYBODY made an online retailer or service website and EVERYONE wanted to invest in that.
Then the market was oversaturated and companies with no right to exist went bankrupt and it was a bloodbath.
Except… not really. Because while the massively overinflated stock market did indeed “downturn” and a LOT of those scam companies went away, the actual fundamental premise of online first companies was a very sound one. I mean… just look at “Cyber Monday” and so forth.
And “AI” will almost definitely go the same route. Because, yeah, LLMs are HORRIBLE for accounting and finance. But they are actually really good for replacing the early career folk who translate earnings into reports. And ML in general is excellent at detecting patterns which can mean potentially billions of dollars in investing. But, like all things, it is about verification and caution. You actually need a human to read that earnings report before you send it to the investors. And you only give your “AI” a small portion of your portfolio. Same as with any team.
In 2011, there was a large flood in Thailand that impacted ~40% of hard drive manufacturing. As a result, hard drives significantly increased in price. This was back when SSDs weren’t mainstream yet.
A year or two later, when manufacturing capacity was restored, prices were essentially back to what they were before the disruption.
Apart from disruptions like that, HDDs, SSDs, and RAM have always been going down in price.
Prices on memory have virtually always gone down, and at a rapid pace.
https://ourworldindata.org/grapher/historical-cost-of-computer-memory-and-storage
What you are describing is something different… that is “close enough” to Moore’s Law for all but the most pedantic.
The (I forget the proper economics term so) base price of RAM/Storage does indeed go down as new processes and economies of scale are developed. But the cost of a “laptop hard drive” remains pretty steady in the sense that a couple hundred MB was enough back in the day but you REALLY want at least 500 gigs now. The price per byte does indeed drop rapidly but the price per “drive” is far more stable (not fully stable due to inflation and how many people are buying them, but within spitting distance).
Its why a good rule of thumb was to always just spend roughly the same on storage during an upgrade and that would result in faster technologies and larger capacity drives and so forth.
That isn’t what is happening with RAM in 2025. A much better comparison is GPUs because… it is the same problem. It is ridiculously high demand from businesses (often startups pouring dump trucks of VC money into their only hope… well, VC money or drug money in the case of miners but they matter a lot less these days) driving this. A quick search didn’t yield an easy graph and I can’t be bothered to go dig through Gamers Nexus’s twelve videos on it, but the price of an “entry level” GPU has drastically changed in the past decade.
But just for two-ish data points?
The last point being what is, by all accounts, going to be the new normal. Barring outside impacts like… RAM going through the roof. Vendors will sell the cards for the ACTUAL MSRP rather than the inflated demand prices. And they will still be considerably more expensive as a result.
All of which is to say… my current card is definitely good enough but having a hard time deciding if I do one “final” upgrade for the decade. But I am an AMD boi so those are at least “reasonable” in terms of price per performance.
people say go back in time to pick the correct lotto number
I say go back in time and sell my 8TB disk for 80 billion
Well, no. Computers until recently (2000s or so) were incompatible with modern protocols required for that much memory.
In 1995 I bought 4 MB of RAM for DM 200, which, adjusted for inflation probably works out to about €200 in today’s money. I’d say, prices have come down quite a bit, since.