RAMmageddon, Part 11: The Bill of Materials Tips Over

In the same quarter, the PC market drops by a fifth while memory makers report the biggest profits in their history. That is no coincidence. It is the same story seen from two sides.

---

From 15 to 40 percent

In an entry-level or mainstream PC, memory and SSD together accounted for about 16 percent of component costs at the start of 2025. According to Omdia, they now account for between 36 and 42 percent in the third quarter of 2026. Over the same period, the bill of materials of a mainstream device rose from 435 to 699 dollars. Even in the premium segment, the memory share has almost tripled, from 8 to 23 percent.

This number captures the thesis of this series. Not because it is surprising, but because it shows that RAMmageddon has now arrived in the bill of materials. Memory is no longer just a component. Memory is the price driver.

And for the first time, we can see what that does to demand.

Two numbers, one quarter

The global PC market collapsed by more than 20 percent in the third quarter of 2026. IDC measures minus 20.1 percent to 62.7 million units, Omdia minus 21.2 percent to 58.1 million. The difference comes from different market definitions; the direction is unambiguous. Notebooks lose 20.6 percent, desktops as much as 23.5 percent. HP is hit particularly hard at minus 31 percent.

In the same quarter, Samsung reports in its preliminary guidance an operating profit of 107.4 trillion won, roughly 80 billion dollars. That is 782 percent more than a year ago and the first quarterly profit by a tech company above 100 trillion won.

Micron closes its fiscal year with 54.2 billion dollars in quarterly revenue. A year earlier, it was 11.3 billion. Gross margin stands at 87 percent. To be fair: the quarter had 14 weeks instead of 13. That does not change the order of magnitude.

The money missing from the PC market has not disappeared. It has moved.

The company that profits from itself and loses to itself

The best picture of RAMmageddon comes from Samsung itself. The semiconductor division earns more than ever from expensive memory. The mobile division, according to analyst estimates, is heading for an operating loss of around 1.5 trillion won, because exactly that memory sits inside its own smartphones.

So the 40 percent bill of materials from the Omdia report is not an abstract industry phenomenon. It runs right through a single company: one hand sends the other the invoice.

For now these are estimates: Samsung will only publish official divisional results on 29 October. But the direction is already clear today.

Apple fits the picture too. The company loses around eleven percent in units but gains market share. Premium buyers are less sensitive to rising component prices. On a 499-euro notebook, on the other hand, 50 euros more for memory changes the entire calculation.

Where the wafers go

If you want to know where the money goes, just look at the other quarterly results:

• Nvidia generates 96.2 billion dollars in revenue in the July quarter, 92.7 percent of it in the data center. Management expects around 70 percent growth for the next fiscal year and calls that supply-constrained itself.
• TSMC is up 54.6 percent in September compared with the same month last year. High-performance computing now accounts for two thirds of revenue.
• Micron has already sold most of its HBM capacity for 2027, with significant price increases.
• Kioxia has allocated its entire NAND production for 2026 and expects the same for 2027.

The shift is most visible in flash. The share of enterprise SSDs in all shipped NAND bits rose from 26 to 48 percent within a year. Almost every second bit goes into the data center.

And new capacity? Samsung and SK hynix are investing mainly in HBM and DRAM, not in flash. Micron wants to produce more DRAM in 2027, but new fabs take years. Samsung and Micron both expect memory to remain tight until 2028.

So the classic PC competes for the same wafers as an AI data center. Micron earns a 90 percent gross margin in its data center business. An office notebook cannot win that competition.

What this means for mid-sized companies

For IT leaders in Germany, Austria and Switzerland, this has very practical consequences:

• Client refresh gets more expensive or worse. Either budgets go up, or devices ship with less RAM and smaller SSDs. Manufacturers have already cut standard configurations from 512 to 256 GB and from 1 TB to 512 GB.
• Buying early paid off for PCs, not necessarily for the data center. Those who bought endpoints in the first half of the year got it right. In the data center it looks different: many systems will easily run for another two years. Buying more hardware now at record prices is not a strategy.
• Capacity becomes a cost question. Every terabyte that sits reserved but unused, or copied multiple times, costs significantly more today than a year ago.

The forecasts give no reason to wait for relief. Omdia expects a further decline of 24 percent for the fourth quarter and another roughly 7 percent for 2027. The third quarter was probably not the bottom.

Modernize instead of buying more

The two years that existing systems in the data center can still carry are not a postponement. They are a window. Now is the right moment for software modernization, standardization, and a serious look at the cloud-native world.

The reason is simple: cloud-native architectures need significantly less footprint in the data center than classic workload applications. Those who consolidate need less hardware. Exactly the hardware that has never been this expensive.

At the same time, the workloads themselves are changing. Only with a cloud-native structure and a modern data structure does it become possible to use AI in a truly meaningful way.

AI, as I have written several times, is not the goal. It is a side benefit. It comes from consolidation and higher efficiency when you invest time and money in data modernization and application modernization.

Small models instead of the big hammer

At the same time, a second argument for your own data center is gaining weight: sovereignty. According to a Bitkom survey, 97 percent of companies in Germany care where their cloud provider’s servers are located. All respondents would prefer Germany as the location.

What applies to data applies even more to AI. Those who run models on their own data want control over where data and models run. Data sovereignty becomes AI sovereignty.

And an insight is taking hold that comes from the few pioneers: the largest model is not the cure-all. Gartner expects that by 2027 organizations will use small, task-specific models at least three times more often than general-purpose large language models. They are more accurate on domain tasks, respond faster and need less compute.

That fits the shortage exactly. A small model needs less GPU, less HBM and less memory. It can run in your own data center without competing for the same resources as the hyperscalers.

The catch: only those who know their workloads and their value chain in detail can tell which tasks suit a small model. That only works with modernized software and clean data structures. Then it becomes visible where a small, specialized model achieves a big effect with little effort.

Instead of using a sledgehammer to crack a nut, you put the right size in the right place. The order stays the same: modernize first, then use AI meaningfully.

And that closes the loop to the PC market. What has to run on endpoints determines what they need to deliver. Those who modernize applications and data also change the requirements for clients, sometimes dramatically. The next refresh should therefore be driven not by the price tag but by the architecture.

Not a market disruption, but a redistribution

The collapse of the PC market is not a short-term outlier. It shows where the center of gravity of the semiconductor industry is shifting: away from the endpoint, toward the AI data center. And this shift is secured by supply contracts through 2028.

Those who do not plan now end up in the DigitalWasteland: with outdated devices, bloated data estates and budgets that do not fit the new price reality.

Those who honestly assess inventory, demand and efficiency now make their next decisions from a position of control rather than pressure.

How are you using the window? Buy more, extend, or modernize?

---

Transparency: The cover image was created with Gemini. I did the research and fact-checking with AI assistance; the theses and analysis are mine. Since 16 September 2026 I have been a Cloud Native Architect at Everpure (formerly Pure Storage), on the Portworx team. The opinions are my own.

Jens Klasen writes in the Future Proof Tech Briefing about the upheaval in enterprise IT. You can find all previous installments in my newsletter and at klasen.ai.