4th Street Bar Hive-Bar

Hive-Bar Powered by Hive beta-fdb5b5b

Community post

AI Is Bad, but It Only Exposed the Cracks: The Hardware Shortage Was Bound to Happen

2025–present global memory supply shortage

https://en.wikipedia.org/wiki/2025–present_global_memory_supply_shortage

AI is gobbling up all the hardware

AI is gobbling up all the hardware available, literally.

Even DDR2, more than 20 years old, is facing price increases.

AI servers swallow memory, SSDs, wafers used for CPUs and GPUs, and even PCBs!

AMD is even bringing back the Ryzen 7 5800X3D, a processor originally launched in 2022 for the AM4 platform which was launched in 2016.

Old hardware has become a pressure release valve for people who can not afford, justify or even find a new platform.

AI companies are buying the future

They (AI datacenters) are reserving wafers, memory, packaging, storage and everything else before those products are even manufactured.

SK Hynix had already announced discussions for the following years HBM supply were complete and demand for its DRAM and NAND products had been secured.

demand for all DRAM and NAND products secured for next year

Micron was similarly negotiating agreements covering most of its future HBM production.

pricing agreements with
almost all customers for a vast majority of our HBM3E supply in calendar 2026

It means the biggest customers are reserving the most important parts of the supply chain years ahead. By the time normal companies and consumers enter the market they are getting the crumbs left of the supply, in the most literal sense as we have to default to older technology since DDR5 and other equally newer tech is all bought out.

Is AI really to blame?

Is AI to blame? Well of course it is, theyare creating the demand. They are pouring absurd amounts of money into hardware and buying capacity faster than manufacturers can expand it. We may even argue it is immoral to buy out unproduced supply years in advance.

But most interestingly is what that tells us about the system underneath because it has happened before.

By the end of the 2010s and into the early 2020s cryptocurrency miners were buying out every GPU they could find. In 2018, demand from miners was already being blamed for graphics card shortages and distorted prices.

When cryptocurrency demand later weakened, GPU prices started falling again.

First it was crypto, now it is AI, tomorrow it could be autonomous vehicles, virtual reality, robotics, or some technology that does not even exist yet, or even some new application or piece of software.

Anything capable of creating a large enough demand spike could have exposed the same weakness AI is exposing.

Did manufacturers not learn a lesson? They did, but I think that they have learned the wrong one.

They have learned that sudden demand can, and often does, disappear. They learned that expanding too quickly will eventually create oversupply and that oversupply destroys prices and profits.

The logic is simple it is wickedness: If you supply your customer they don't need to buy. It is obvious and stupid but true, if someone wants to buy somehting and you sell, they will buy and you won't have someone wanting to buy. The "want to buy" is what spikes value.

It is a total degeneration of supply chain and market dynamics. We thought that if you sell something you want to sell and get your products in the hands of your customers, as many products as possible in the hands of as many customers. But no, it is more profitable to get some products in the hands of a few customers, this way you ensure you will have someone wanting to buy from you tomorrow because they couldn't today.

Three companies carry almost the entire memory market

The fact that almost all global DRAM production sits on the shoulders of a few companies was a mistake.

According to current marketshare estimates Samsung, SK hynix, and Micron together control nearly 90% of the global DRAM market by revenue.

Three companies, all of which have been investigated for cartel-like pricing and supplying schemes and that now carry almost the entire global market for one of the most important components in modern technology.

CXMT, a Chinese memory manufacturer, has emerged as a fourth major DRAM manufacturer. Its DDR5 chips are beginning to appear in mainstream products including Corsair memory modules.

A fourth serious manufacturer means more production, more competition, and potentially more resilience.

But one new manufacturer cannot undo decades of concentration overnight, specially when it has such a low market share and production capacity. China may be creating another pillar, but the entire building is still balanced on three enormous columns, balancing the load will take years, if not decades. Their strategy of starving their customers today so you will still have buyers tomorro will keep on working as planned.

This is a supply chain problem

Modern production is obsessed with being lean, efficient and almost stock-free. Companies minimize inventory, remove supposedly idle capacity, and depend on just-in-time deliveries, all that under stable conditions is extremely efficient and can, in fact, lower prices.

Running supply chains without enough inventory, spare capacity, supplier diversity, or redundancy turns efficiency into fragility and that is the price for cheaper production and higher margins.

Manufacturers can increase production but increasing production enough to meet a sudden spike creates another risk: by the time the new capacity is ready, the customers may already have what they need, creating oversupply and/or overcapacity and that is considered bad business.

We have already watched manufacturers cut production

During the previous downturn, memory manufacturers deliberately reduced production because there was too much supply and prices were collapsing.

In 2022 Micron announced that it would reduce memory-chip supply and cut capital spending to clear excess inventory

In 2023, Samsung also announced a meaningful reduction in memory production after its quarterly profit collapsed.

We are seeing the aftermath of such decisions. From the perspective of an individual company this behavior is rational becayse nobody wants to spend billions building a factory that may become unnecessary before it opens.

From the perspective of the global economy though, it creates a system that is permanently one large demand spike away from a crisis. It is not the first time this happens and won't be the last precisely because it works (for the manufacturers).

Shortages are profitable

The (production) system is not designed to produce as much hardware as humanity might possibly need, it is not designed to expand production, cut prices and give us abundance as one may idealistically think. It is all designed to produce hardware profitably while minimizing idle factories, unsold inventory, and financial risk.

Then something like AI or crypto happens and guess who takes the short end of the stick?

AI tipped the ship

AI tipped the ship but the issue we are seeing was bound to appear eventually. If it had not been AI or cryptocurrency mining, it would have been some other technology or trend capable of pushing demand beyond the tipping point.

Companies have repeatedly cut production after periods of oversupply and when they expand again they do it cautiously.

Manufacturers are finally scaling up again but they can scale slowly and still fail to meet demand for years, that is by design.

AI made the cracks impossible to ignore

The world is fucked up and our supply chains were not built to supply us, they were built to supply us profitably. AI did not create that system and neither did crypto when it was their turn to gobble up everything.

AI is bad but the hardware shortage was bound to happen.

This is a repeated cycle. It has happened before and it will happen again.

627 upvotes $2.08

Replies (85)

Review before signing

Posting as . Signing with . Keychain permission: Posting. Hive Keychain will ask you to approve this action next. 


  

    Technical details
    

      Operation fingerprint: