RAM makers are taking on massive debt to keep up with AI's chip appetite

[Alfonso Maruccia]

The big picture: AI infrastructure demand is no longer just affecting enterprise hardware. The surge in HBM and server DRAM demand is now reshaping the broader memory market, pushing up pricing for consumer SSDs and RAM while forcing downstream companies to aggressively secure inventory before costs rise even further.

Correction (May 19): This article has been updated for clarity and accuracy regarding memory pricing trends and certain reported revenue growth figures and framing cited from translated industry reports.

AI-fueled memory shortages are becoming so severe that several Taiwanese memory module makers are now borrowing hundreds of millions of dollars just to secure enough DRAM and NAND inventory to keep up with demand.

According to Taiwan's Commercial Times, companies including Adata, TeamGroup, Apacer, Innodisk, Transcend Information, and Silicon Power are collectively raising more than NT$28 billion, or roughly $880 million, through convertible bonds, syndicated loans, and private share placements to fund memory chip purchases.

The largest fundraising effort comes from Adata, which reportedly completed a NT$2 billion convertible bond issuance, secured NT$12 billion in bank loans, and is preparing a 30 million-share private placement. GoldKey Technology raised NT$4.5 billion through a mix of bonds and loans, while TeamGroup and Apacer completed NT$2 billion and NT$1 billion convertible bond issuances, respectively. Innodisk and Transcend are each planning NT$3 billion offerings, while Silicon Power is preparing a smaller NT$500 million issuance.

The situation highlights just how distorted the memory market has become during the current AI boom. These companies are not struggling financially. In fact, several are reporting record revenue growth.

Adata's March revenue exceeded NT$10 billion for the first time, while its Q1 revenue more than doubled year over year to NT$26.11 billion. TeamGroup also reported a sharp revenue increase in March, though some translated reports describing triple-digit sequential growth figures may require additional clarification.

Despite those gains, maintaining sufficient inventory has become increasingly expensive as DRAM and NAND contract pricing continues to surge.

Market research firm TrendForce estimates that conventional DRAM contract prices rose between 90% and 95% quarter over quarter during Q1 2026, with another 58% to 63% increase expected in Q2. NAND flash pricing has also climbed sharply, rising roughly 60% in Q1, while additional increases are expected throughout the second quarter.

More recent TrendForce projections from early May suggest mobile DRAM pricing could rise even further as suppliers including Samsung Electronics, Micron Technology, and SK hynix continue prioritizing high-margin server DRAM and HBM production tied to AI accelerators and cloud infrastructure deployments.

That shift has left module makers with limited influence over supply allocation. Unlike Samsung, Micron, or SK hynix, companies such as Adata and TeamGroup do not manufacture DRAM or NAND chips themselves. Instead, they purchase finished memory components and assemble them into consumer products including DDR5 memory kits, SSDs, USB drives, and industrial storage solutions.

As a result, securing inventory early has become one of the few tools available to protect margins and maintain product availability during the ongoing supply crunch. The strategy also reflects growing concern that meaningful new fab capacity may not arrive until 2027 or later, potentially extending elevated pricing across the consumer memory market well into the future.

Permalink to story:

RAM makers are taking on massive debt to keep up with AI's chip appetite

 

[Hecate91]

Memory suppliers are forced to take out loans to resupply the market, what the heck? This is wrong on so many levels, the RAM cartel should be investigated for collusion and price fixing again.
If the smaller memory suppliers like Adata or Teamgroup were having record profits, you'd think they wouldn't need to borrow money, it simply makes no sense.
The AI demand, AKA the circular money scheme being passed around from companies like Nvidia and OpenAI is expected to keep going until at least 2028, even then the market would take time to recover.

 

[ScottSoapbox]

Memory suppliers are forced to take out loans to resupply the market, what the heck? This is wrong on so many levels, the RAM cartel should be investigated for collusion and price fixing again.
If the smaller memory suppliers like Adata or Teamgroup were having record profits, you'd think they wouldn't need to borrow money, it simply makes no sense.
The AI demand, AKA the circular money scheme being passed around from companies like Nvidia and OpenAI is expected to keep going until at least 2028, even then the market would take time to recover.

I know it is hard to believe things that don't fit one's preconceived ideas, but needing to borrow money to expand to meet demand is not proof that they are making even more collusive profits than you thought, but the quite opposite.

It's like thinking Dell needing to spend more money buying RAM for it's laptops shows it's colluding on laptop prices. The inputs are getting more expensive - because AI demand is breaking so many things.

Further, manufacturing for cutting edge tech is expensive. IF demand continues then yes the very expensive machine today will pay for itself over the next X years, but that doesn't show collusion. That's how manufacturing investment works as a basic principle.

 

[godrilla]

On other hand if they didn't borrow the money which is being offered by banks as an opportunity for almost short term guaranteed growth would be a fiduciary disservice to the share holders imo.
The shareholders will quickly remind those suits via better opportunities just like today's Seagate stock sell off by not having an expansion roadmap.

 

[pnntmp]

The disinformation per unit of text ratio achieved in this article is amazing!
State of the art!
It should be studied in propaganda schools!

Everything positive is presented as negative.
In reality, RAM makers are drowning in profits.

 

[fadingfool]

Borrowing money to invest in the production capabilities of your plant to meet unprecedented demand is what we call that weird concept called "Business".

 

[user556]

The pop and crash is gonna be loud and gruesome!

 

[Inthenstus]

The disinformation per unit of text ratio achieved in this article is amazing!
State of the art!
It should be studied in propaganda schools!

Everything positive is presented as negative.
In reality, RAM makers are drowning in profits.

Yeah but how are they going to make a profit if we don’t think there is a shortage? /s

 

[bviktor]

Clickbait title - do better Techspot! And don’t edit my post because you don’t like what I say!

They are not DROWNING in debt! They are borrowing money so that they can PROFIT from the insane demand for their products!

They’ll (and maybe we’ll) find out if they are “drowning” after the demand lets up in a few years…

Yes, that's how debt works, it doesn't hit you instantly, thanks for the incredible insight.

 

[Squid Surprise]

Yes, that's how debt works, it doesn't hit you instantly, thanks for the incredible insight.

It's NOT an insight... and this story has a clickbait title (thanks for quoting me, as they deleted my post). Companies go in debt all the time - some "drown" and go under... some use the capital they borrowed to make more profits - and repay their debt...

This story is saying that these memory companies are "drowning" and implies that they will go under... this is almost certainly NOT the case!! With record profits in this sector, there's a REASON these companies are borrowing so much money! They're not 1diots! They see the chance to make even more money in the future - otherwise they wouldn't be borrowing!

 

[Julio Franco]

Correction (May 19): This article has been updated for clarity and accuracy regarding memory pricing trends and certain reported revenue growth figures and framing cited from translated industry reports.

 

[Rapob]

The pop and crash is gonna be loud and gruesome!

There won't be any crash, as we need massive amount of chips in the next 10 years, probably forever from here. We also need 10 times the energy.

Even with all the upcoming fabrication facilities, we will lack chip production.

We are slowly prepairing to leave the tier 0 civilization status. Go read about the Kardashev scale. Humans won't evolve much without AI going forward. We maxed out.

The future for humans is tech/AI. AI needs chips and energy. We will evolve more in the next 25 years than we did in the last 50 years. We are not advanced.

When it comes to AI, you have seen nothing yet. This is early days of AI. Very early days. And demand won't stop from here, only grow. Agentic AI will need vastly more compute and power and this is when you will see what AI is truly capable of.

Boggles my mind that people think market is going to crash year after year. While completely missing out on cashing in. Seriously, get on board. Or get left behind but then you will complain about high prices for the rest of your life. Mark my words.

 

[Vanderlinde]

Instead of ever expanding with RAM; why not optimize the current code set or the way AI is storing things, to start using less ram? I'm pretty sure you could get so much more out of a base set of hardware, if you just optimize it rather then smashing another few billions of USD to store extreme large datasets.

 

[Theinsanegamer]

Instead of ever expanding with RAM; why not optimize the current code set or the way AI is storing things, to start using less ram? I'm pretty sure you could get so much more out of a base set of hardware, if you just optimize it rather then smashing another few billions of USD to store extreme large datasets.

Because it's really easy to say "JuSt OpTiMiZe" but it's really hard to actually accomplish. If you're working with dozens of gigabytes of data.....you gotta put that data somewhere, you have to decompress it eventually.

There won't be any crash, as we need massive amount of chips in the next 10 years, probably forever from here. We also need 10 times the energy.

Even with all the upcoming fabrication facilities, we will lack chip production.

We are slowly prepairing to leave the tier 0 civilization status. Go read about the Kardashev scale. Humans won't evolve much without AI going forward. We maxed out.

The future for humans is tech/AI. AI needs chips and energy. We will evolve more in the next 25 years than we did in the last 50 years. We are not advanced.

When it comes to AI, you have seen nothing yet. This is early days of AI. Very early days. And demand won't stop from here, only grow. Agentic AI will need vastly more compute and power and this is when you will see what AI is truly capable of.

Boggles my mind that people think market is going to crash year after year. While completely missing out on cashing in. Seriously, get on board. Or get left behind but then you will complain about high prices for the rest of your life. Mark my words.

The exact same arguments were used for crypto, and NFTs, look where those ended up.

Here are some fact checks:
Half of data center projects have either been delayed or cancelled outright

Nvidia's inventory has quadrupled in size despite all this "demand"

AI companies like openAI have yet to demonstrate any sort of path to sustainable profitability

And lastly:
LLMs are not new. They have existed for decades. They are not intelligent and they are not that capable. AI has its uses, certainly, but as we can see, it is also massively expensive to maintain and its scope is rather limited.

All the tech from the dotcom bubble was useful. All the servers, networking equipment, and visions of the digital future came to fruition. Yet the bubble did pop, and it was the third largest financial failure in US history that derailed retirements for millions of Americans and drove multiple large, profitable companies bankrupt.

Just because a tech is useful does not mean that a bubble does not exist.

 

[Rapob]

Because it's really easy to say "JuSt OpTiMiZe" but it's really hard to actually accomplish. If you're working with dozens of gigabytes of data.....you gotta put that data somewhere, you have to decompress it eventually.

The exact same arguments were used for crypto, and NFTs, look where those ended up.

Here are some fact checks:
Half of data center projects have either been delayed or cancelled outright

Nvidia's inventory has quadrupled in size despite all this "demand"

AI companies like openAI have yet to demonstrate any sort of path to sustainable profitability

And lastly:
LLMs are not new. They have existed for decades. They are not intelligent and they are not that capable. AI has its uses, certainly, but as we can see, it is also massively expensive to maintain and its scope is rather limited.

All the tech from the dotcom bubble was useful. All the servers, networking equipment, and visions of the digital future came to fruition. Yet the bubble did pop, and it was the third largest financial failure in US history that derailed retirements for millions of Americans and drove multiple large, profitable companies bankrupt.

Just because a tech is useful does not mean that a bubble does not exist.

Look where crypto ended up? BTC and ETH, the only two that matters, literally peaked late last year and will peak even higher in the coming years. 200-250K USD per bitcoin within a few years is my prediction from here.

NFTs was sh1t from day one, no comparison. No-one ever cared. Garbage.

This has nothing to do with Tech/AI tho. The demand is huge here and won't get lower anytime soon (next 5-10 years). Supply is much lower than demand. Production is the bottleneck.

You are clueless about the data centers, I do that for a living. We pretty much have orders for the next 5 years lined up and people wait for months and years right now, depending on size.

I seriously laugh when people talk about a crash in tech/AI. Always coming from people who missed the train or don't know anything about the sector. Absolutely no sign of a bubble. Demand is huge and will only grow. Production is the main problem. Hence why they build fabs left and right and even when all these stand ready, we will still lack production. That is how insane the demand is.

 

[Endymio]

Yeah but how are they going to make a profit if we don’t think there is a shortage? /s

Memory makers will garner large profits due to datacenter and AI purchases, no matter whether or not a few video-gamers reject the reality of a global shortage of RAM.

Instead of ever expanding with RAM; why not optimize the current code set or the way AI is storing things, to start using less ram?

False dilemma fallacy; they're doing both. But the usage of AI and proliferation of new AI models is far outpacing gains on the software side.

Memory suppliers are forced to take out loans to resupply the market, what the heck? This is wrong on so many levels, the RAM cartel should be investigated for collusion and price fixing again.

A basic understanding of economics should be a requirement to vote in a liberal democracy, else one is so easily gaslit into statements like the one above. When demand for a product skyrockets much faster than supply can keep pace, there are only two possible choices: either prices rise to control that demand, or the shelves run bare. No matter how hard one whines and stamps their feet, one cannot wish several cutting-edge multi-billion dollar chip fabs into existence overnight.

 

[agentnate009]

On other hand if they didn't borrow the money which is being offered by banks as an opportunity for almost short term guaranteed growth would be a fiduciary disservice to the share holders imo.
The shareholders will quickly remind those suits via better opportunities just like today's Seagate stock sell off by not having an expansion roadmap.

Investors trade on emotions quite a lot! From Seagate's perspective, expanding capacity for HDD/SSD would be a waste of money once AI boom falters. They have to consider the long term, you just don't build uber expensive chip factories for a several year boom that may fizzle out not long after you get said uber expensive factory up and running.

 

[Squid Surprise]

Investors trade on emotions quite a lot! From Seagate's perspective, expanding capacity for HDD/SSD would be a waste of money once AI boom falters. They have to consider the long term, you just don't build uber expensive chip factories for a several year boom that may fizzle out not long after you get said uber expensive factory up and running.

HDD factories vs chip factories are apples and oranges... HDD tech is obsolete - but because of how expensive SSDs are, they remain the cost-effective way to have mass storage. Obviously, a company would be hesitant to commit long term to making tons of them (but I'd wager they'll stick around for at least 10 years).

Chip factories are different - we'll ALWAYS need chips! What gets manufactured in each fab changes as tech does - but they are far easier to commit to long term.

 

[Hecate91]

Economics would dictate that more ram factories be built to meet demand, however that isn't happening because the ram cartel wants to keep prices inflated, while they're also concerned that AI is a bubble. When the ram producers think AI is a bubble it should be obvious that AI isn't going to keep the billions being passed around or VC's funding AI companies forever.

Look where crypto ended up? BTC and ETH, the only two that matters, literally peaked late last year and will peak even higher in the coming years. 200-250K USD per bitcoin within a few years is my prediction from here.

NFTs was sh1t from day one, no comparison. No-one ever cared. Garbage.

This has nothing to do with Tech/AI tho. The demand is huge here and won't get lower anytime soon (next 5-10 years). Supply is much lower than demand. Production is the bottleneck.

You are clueless about the data centers, I do that for a living. We pretty much have orders for the next 5 years lined up and people wait for months and years right now, depending on size.

I seriously laugh when people talk about a crash in tech/AI. Always coming from people who missed the train or don't know anything about the sector. Absolutely no sign of a bubble. Demand is huge and will only grow. Production is the main problem. Hence why they build fabs left and right and even when all these stand ready, we will still lack production. That is how insane the demand is.

Right, look at what happened to crypto, only the crypto bros care about it anymore.
Sure AI does have its uses, but the valid uses it has don't need datacenters consuming more electricity than an entire state.
The whole AI boom is unsustainable at the rate that the AI tech bros think things can progress, in reality, AI isn't smart and building more server farms isn't making LLMs any smarter.
Also how many of those orders are actually getting fulfilled?
Over half of datacenters being built are sitting empty because the chips to power servers, and the power grid infrastructure doesn't exist, it's a massive scam at this point because these datacenters simply aren't going to pay off.

 

[Axeia]

This has nothing to do with Tech/AI tho. The demand is huge here and won't get lower anytime soon (next 5-10 years). Supply is much lower than demand. Production is the bottleneck.

Considering the amount of news posts about hyperscalers having more cards than data centers and data centers being build faster than the energy providers/net can be scaled up chip production seems to have somewhat stopped being the one singular bottleneck.

Absolutely no sign of a bubble. Demand is huge and will only grow. Production is the main problem. Hence why they build fabs left and right and even when all these stand ready, we will still lack production. That is how insane the demand is.

Ehhhh
Absolutely no sign of a bubble is something you won't even hear Sam Altman say. In fact, even he stated it's a bubble.
We won't know for sure until/if that that demand vanishes overnight if it is a bubble or not but I'm definitely leaning towards yes. Every study points out that the amount of companies actually making profit after implementing AI is actually stuck with higher costs with often times little to no benefit - or worse a decline in quality/productivity.

We're very much still in the all we have is a hammer and everything looks like a nail phase. AI is being thrown willy nilly at every sector in every possible form to see what works and the answer so far overwhelmingly seems to be - it doesn't.
I'm very much expecting a strong analog to the dot com bubble where eventually a few companies come out on top but a great many go under (Google will be fine, but will OpenAI? Will the companies that offer little more than a thin wrapper around existing models?). We'd need actual AI to take the baton and keep running because LLMs aren't the solution to everything.

 

[Squid Surprise]

Right, look at what happened to crypto, only the crypto bros care about it anymore.

really? So why is a Bitcoin trading at over $75,000 today?
Must be a lot of “crypto bros” out there.

When you spout ignorant nonsense like that, it’s hard to take anything you say seriously.

 

[Hecate91]

really? So why is a Bitcoin trading at over $75,000 today?
Must be a lot of “crypto bros” out there.

When you spout ignorant nonsense like that, it’s hard to take anything you say seriously.

Found the crypto bro.
High trading prices doesn't mean it's not massively overvalued, or the average person cares about it. I can't go buy food or gas with a bitcoin, it still has to be converted to real currency.
And look at where all of those other coins went after the crypto crash, when AI inevitably crashes there might be a few "AI companies" left, and they likely won't be giving out LLM prompts out for free.

 

[Squid Surprise]

Found the crypto bro.
High trading prices doesn't mean it's not massively overvalued, or the average person cares about it. I can't go buy food or gas with a bitcoin, it still has to be converted to real currency.
And look at where all of those other coins went after the crypto crash, when AI inevitably crashes there might be a few "AI companies" left, and they likely won't get giving out LLM prompts out for free.

Actually, high trading prices mean it’s valued exactly what it’s worth… that’s how markets work… and I can’t buy something with RAM either - doesn’t mean it isn’t worth anything…

Crypto will last longer than any national currency. Every nation/empire eventually fails - but the internet is gonna be around awhile… try buying something with Carthaginian currency nowadays…

 

[onebignerd]

Going into debt for centers that have yet to materialize, there is insufficient power generation and water for more data centers.

When AI gold rush dies, it will be a blood bath!!

 

[Squid Surprise]

Going into debt for centers that have yet to materialize, there is insufficient power generation and water for more data centers.

Many have… and there’s plenty of power generation and water… don’t believe the nonsense you get in shock media…

When AI gold rush dies, it will be a blood bath!!

When the AI rush ends, it will be because we have restored our supply chain and we’ll be able to construct as needed - and we’ll always need