Hello can someone explain why ai made ram prices higher?

[-]

9okm@reddit

AI uses a lot of ram. It's not that complicated.

[-]

meicz@reddit (OP)

Ahh it’s that simple. Was just reading other stuff and they made it super complicated, or maybe I’m just an idiot 🤔

[-]

yalocalsnack@reddit

yeah ur just an idiot cuz google is free now stfu

[-]

The global/economic constraints that make RAM a limited resource are complicated. The fact that AI needs a lot of it, thus reducing RAM supply, is the simple part. The higher the demand for AI data centers, the higher demand for RAM, thus reducing the supply of RAM, and resulting in a much higher price for consumer RAM.

[-]

HWKII@reddit

Rising demand doesn’t reduce supply, it simply consumes more of the available supply.

[-]

SpoopyJD@reddit

…resulting in less supply…

[-]

HWKII@reddit

No. Supply is what’s provided by the market, it’s a verb. Demand does not reduce Supply, rising Demand invites more Supply.

[-]

SpoopyJD@reddit

But there isn’t more supply! It’s a limited supply! That’s the point! This isn’t a hypothetical from your Econ 101 class man, they can’t produce enough ram to meet demand. Thus, price go up.

[-]

dragonblade_94@reddit

higher price for consumer RAM

It should be said that even sever/enterprise focused memory is seeing a huge price spike, due to the fact the datacenter buyers are bidding well over market to secure their inventory.

[-]

9okm@reddit

It also helps to realize that there is only so much ram that can be produced. It's supply is more or less finite - like gold. So when suddenly one group (AI companies) really really really really really want as much of it as possible, it becomes more expensive.

[-]

WintersDoomsday@reddit

Low supply without lower demand causes high prices....not hard to grasp honestly. But people want to defend their garbage enjoyment of social media posts showing stupid AI videos of the cast of full house then meeting their older selves......cause that's worth tripling the price of computer components and water usage.

[-]

lordhooha@reddit

That’s not the AI that’s using all this ram.

[-]

Global-Awareness6961@reddit

This is the big reason. If computer part manufacturers could be pumping out more ram to meet ai demand, they absolutely would, because they would be able to make even more money than theu are now. Its just that the infrastructure and machinery needed to make complicated, intricate computer parts costs literally billions of dollars of investment and many years to set up and get running. That is if everything goes smoothly, at least. If construction faces issues or anything then it could end up costing a lot more and taking a lot more time.

[-]

9okm@reddit

I can't recall where I heard it... but I believe many memory manufacturers are actually intentionally NOT working to increase capacity. It's kinda nice when they know they can sell 100% of their capacity at whatever price they want. Plus, with the fears of a bubble bursting, if they poured millions into infrastructure that is no longer needed... that'd be a big problem.

[-]

Global-Awareness6961@reddit

Yeah I should've said that to.

If they could meet the demand they definitely would, but the only way for them to do so is to gamble on whether or not ai is a financial bubble right now and if the demand will survive for long enough for new production to pay itself off- and anybody with a brain not blinded by ai hype can tell you that all of it absolutely is overvalued right now.

Why would a company like Samsung or TSMC invest in new production when theres a 95% chance it'll cause them to be a few billion dollars in the hole by the end of 2028.

[-]

lordhooha@reddit

Maybe so

[-]

footpole@reddit

Both can be true

[-]

captainstormy@reddit

It's just supply and demand at the end of the day. There are only a few companies that make ram and they only have so much capacity.

Demand has skyrocketed because our current AI models need a stupid amount of ram. Data centers can't get their hands on enough and have started ordering literally as much as the suppliers will sell them.

I'm honestly surprised RAM is even available for consumers at this point.

[-]

Accomplished_Emu_658@reddit

They make it sound complicated intentional in articles intentionally to be misleading.

[-]

Arsys_@reddit

on top of that they got multiple data centers for AI, so a big structure of only PCs everywhere means more hardware

[-]

edgeofsanity76@reddit

It's not that, it's more the type of ram they need

[-]

PantsMcGillicuddy@reddit

So...AI uses a lot of RAM that's the same type people want for other products. Or shorter...AI uses a lot of ram.

[-]

No_Life_1315@reddit

saw this and had to sigh because my own pc upgrade is on hold right now thanks to these prices.

i never really grasped just how heavy ai models were until recently. i’m a few months into an internship at Lifewood Data Technology and i’m still lowkey overwhelmed by the sheer size of the datasets we work with. we process literal billions of text tokens and massive image batches just to get the data ready for the models. seeing how much computing power is needed just on our end for data validation and pipeline formatting makes my head spin. if prepping the raw data takes this much juice, the actual model training must eat ram like absolutely crazy.

guess everyday pc builders are just competing with massive server farms for memory now. rip to our wallets lol.

[-]

psimwork@reddit

There's two factors in play: increased demand, and reduced production translating into reduced supply.

AI datacenters use DDR4/DDR5, but not in huge amounts. So that results in (moderately) increased demand. If supply remained steady with regards to production of DDR4/DDR5, there'd be an increase in RAM prices, but not to the extent that we've seen it currently.

The bigger issue is that datacenters need HUGE amounts of a memory type called HBM (high bandwidth memory). This is memory that doesn't have a huge capacity, but rather EXTREMELY high bandwidth.

The problem is that HBM sells for a LOT higher price than DDR4/DDR5, but there isn't usually much demand for it. But now, AI datacenters are basically saying, "produce all you want - we'll buy all of it. Guaranteed." And the machines that make DDR4/DDR5 are the same machines that make HBM.

So now let's say you're a RAM manufacturer. There's only like 5 or so in the entire world. You can't just increase your production capacity, because that would literally cost billions. Instead, you have a certain number of machines that can produce [x] wafers per-day. Are you going to produce [x] wafers of DDR4/DDR5? Or are you going to produce [x] wafers of HBM, which you can sell for a much higher price?

The RAM manufacturers have largely all changed to producing HBM, meaning that there's very little DDR4/DDR5 being produced. Demand hasn't gone down (in-fact, it's increased slightly), so the price rises accordingly.

[-]

meicz@reddit (OP)

So what I’ve gotten from this is that they ditched making ddr4 and 5 because hbm is better and more useful for these ai companies resulting in regular users not being able to get ram? I hope I’m correct I have a hard time reading 😓

[-]

psimwork@reddit

Think of it this way - you run a restaurant. You can produce 100 plates of food in a day. You are doing well making plates of burgers and fries, and people are buying it. But then Sam Altman comes in and says, "I don't like burgers and fries. I want Steak Tartare. And my friends all love it too. I want you to produce nothing but Steak Tartare." You ordinarily sell your burgers and fries for $10 per plate. But you tell Sam that Steak Tartare will run him $100 per plate. He says, "I don't care. It's what I want."

Now - are you going to continue to make burgers and fries? Or are you going to make the product that sells for 10x the price?

In the meantime, the people that were buying your burgers and fries still want to buy their food, so they have to go elsewhere. The increased demand elsewhere causes a rise in prices.

[-]

Instruction_Boring@reddit

sam altman is the asshole, find him and stop

[-]

9okm@reddit

Yup.

[-]

Sleddoggamer@reddit

AI uses a lot of ram and processing power. The companies went as far as to take out loans to buy out all the ram before it left the factories so there's a shortage, making it much kore valuable then it used too be

The same thing goes for storage, and GPUs have been in the rise too

[-]

meicz@reddit (OP)

Man I’m cooked if I wanna upgrade my pc 😭💔

[-]

AcanthaceaeClean5921@reddit

Absolutely. The prices ain't going down for several years

[-]

psimwork@reddit

There may be a slight silver lining to this. Game makers ain't gonna stop making games. And they make their games based on constantly-moving-forward hardware specs.

Recently, the concept of optimization seems to have taken a big backseat, preferring to slap band-aid solutions like DLSS onto things to make it seem like the performance is better than it actually is.

There's a possibility, albeit somewhat remote, that game makers will start bringing back a culture of optimization to their games if the pace of hardware upgrades slows to a crawl.

[-]

WintersDoomsday@reddit

You act like prices ever go down for anything other than TV's

[-]

Zaphod392@reddit

Yea, wouldnt suggest any upgrades till like 2028 if possible

[-]

Ok_Perspective_7978@reddit

It's partially because of chip shortages too. In combination with some manufacturers stating they will no longer be making consumer grade ram to focus on the ram components needed for AI.

It's all the perfect shit storm centered around AI

[-]

Boys4Ever@reddit

Supply and demand.

[-]

DandyGoon@reddit

AI needs a lot of ram for inference and context. There is a limited amount of ram produced in the world by only 3 OEMs. Companies have purchased and pre ordered future batches of ram up to 2030.

OEMs that make ram and other silicon components refuse to expand because they stand to make more money if they don’t.

Companies believe it’a a race to the top, but it’s actually a race to the bottom for society in every aspect.

[-]

SuperBAMF007@reddit

Supply/demand.

1 AI datacenters require lots of RAM 2 AI companies want to buy RAM = demand for RAM goes up 3 Supply for RAM cannot meet the demand for RAM = Price for RAM must go up 4 Price goes up, less people can afford it and try to buy it, demand goes down 4 More RAM for datacenters

Theoretically, if AI datacenters no longer need more RAM because they’re all built already, but RAM OEMs still make the same amount as they had (or even just the stockpile of RAM is enormous), the price of RAM will plummet.

And then tangentially, because GPUs have VRAM and modern SSDs use much of the same technology as RAM, that’s why GPUs and SSD prices are going up at the same time (albeit to a lesser degree)

[-]

3ofUsDeez@reddit

Usually .. when stuff sells well, demand is high .. prices used to go down

But with GPUs and the supposed mining craze? .. nope .. we get gouged because people will pay stupid prices

Now it's happening with RAM.. Ai companies buying up everything .. so sales are good right? RAM companies getting their products sold and they are making bank right?

That is correct! So you know what Mr. And Mrs. Customer..? F.U. , that's what... I'm going to gouge your eyes out because you'll pay those inflated prices

THEN .. when supply can start meeting demand a little bit, I'll come down from our nonsensical 500+% mark up and drop down to "just" 200-300% mark up so you'll think you're getting a deal .. just like my good buddies over at their GPU plants did..

[-]

Oceanz08@reddit

because Corporations are greedy. Not to mention, Ai is a massive bubble that every company is trying to get the most out of Ai for Profit reasons

[-]

Vashsinn@reddit

Imagine you are hungry and ask your mom for some milkies.

Your brother who is older has eaten all the milk and is malnourished so your mom says he needs more milkies than you.

That's it. Replace milkies with ram, replace your older fatter malnourished brother with AI.

[-]

MinimumMarsupial6782@reddit

If you're a new born then - "Gooo gooo gaaa gaaa!! Yaaaay weeeeeeee!!!"

[-]

Zaphod392@reddit

Ai companies literally bought ALL the RAM that is out there and even the future ones for the next year or two. There is no supply but the demand is high.

[-]

Ok_University_5352@reddit

From a tweet that had gone viral like a month ago, but can't share a screenshot:

"The reason why RAM has become four times more expensive is that a huge amount of RAM that has not vet been produced was purchased with non-existent money to be installed in GPUs that also have not vet been produced, in order to place them in data centers that have not yet been built, powered by infrastructure that may never appear, to satisfy demand that does not actually exist and to obtain profit that is mathematically impossible."

Long story short, a massive contract tied up ram for the next 3-4 years, long before even being produced, so that data centers for OpenAI and others can have it.

Obviously buying in massive bulk quantities can give better margins and they will also pay higher premiums on ram as opposed to a consumer. Hence why Micron has exited the consumer market for ram.

This is also a reason why gpus are starting to increase in prices again and storage drives have shot up 2-4x the price as well.

[-]

Big-Pressure-918@reddit

AI is powered by these giant ass buildings known as "data centers."

Inside these buildings are basically thousands of computers (not exactly like but similar to a personal computer or laptop). These computers are mounted in racks and each one requires its own RAM to function properly.

When a company like open AI has plans to open 10 new data centers across the globe within a few months, they contact Samsung, SK Hynix, and Micron to order a huge amount of their RAM production ahead of time to go into these data centers.

When they buy in bulk like this, even if the RAM manufacturers heavily discount the RAM for Open AI or similar companies, it's still a much more profitable way to sell RAM because of the time it takes to move such a large amount of volume. A data centers worth of RAM being sold in the consumer market would take significantly longer to move all of that inventory.

So basically, it's more profitable for RAM manufacturers to sell in bulk to companies like Open AI who need it for data centers. This has drastically reduced the supply of RAM in the consumer market, while demand has remained the same. This causes (more like allows) the RAM to drastically increase in price.

[-]

Naerven@reddit

Because AI computers use computer hardware on a very large scale. Since things like system memory have a finite production limit that was already being used every year and AI companies will pay more than consumers they reserved some 70% of the supply through 2027.

[-]

TechnoGMNG589@reddit

As a more indepth explanation, Ai data centers arent taking ddr5 gaming ram, their taking the COMPONENTS to make ddr5 ram, and making the ram compatible for ai datacenters, of which the money they make from that is to such a degree its not worth it for companies to make ram for us.

[-]

MacintoshEddie@reddit

Datacenters use massive amounts of hardware.

Imagine a computer that has 50TB of storage, and 256GB of RAM, and now you're barely scraping what they're dealing with, multiplied by several hundred.

[-]

R3D_T1G3R@reddit

Basically RAM goes into GPUs, known as VRAM.

LLMs are kinda huge, with smaller ones starting at some GB, and larger state of the art models being over 1TB in size.

For those to operate fast enough you'll need to load them into memory.

Nvidia has funky adapters to link their data center GPUs together so you can run massive models in the VRAM of your GPU.

If your VRAM isn't enough you put parts of the model into the RAM.

SSDs are also more expensive because they occasionally contain some DRAM, usually 1-4GB, and they're also used by data centers to store those models so they can be loaded a bit quicker.

While HDDs are also more expensive because they're being used to store tons of training data around the world.

[-]

SignalButterscotch73@reddit

They bought it all, meaning there's hardly any left for normal users. Less supply rises the price.

[-]

Combatical@reddit

In fact they bought so much that they bought ram that wasnt even created yet physically.

[-]

ddhh@reddit

More AI stuff = more servers, those eat tons of RAM, demand shot up fast

[-]

edgeofsanity76@reddit

They don't want your DDR5. They want the production capacity for HBM production. High bandwidth memory for servers and specialist GPUs.

There are only a handful of memory manufacturers, so they have bribed them to switch to HBM production. This is why micron have dropped consumer products and gone all in on HBM. Cash.

[-]

CraziestCanuk@reddit

There is only a certain amount of RAM physically able to be produced each year and data centers have bought (literally) all of it for 2026.. They can buy in bulk and are less hassle to deal with than regular consumers so they get priority, and consumers end up with the tiny bit of "extra" that can be squeezed out of the production lines (or stock from previous years etc..)

[-]

fc_dean@reddit

OpenAI (ChatGPT) purchased almost all RAM production capacity for the whole 2026 and 2027, leaving nothing for consumers.

Low (or in this case almost none) supply + the usual demands = Extremely higher prices.

[-]

meicz@reddit (OP)

Thank you👍👍