Chip Crisis 2.0: AI Is Not Consuming All Chips, Only the Most Valuable Ones

On Friday afternoon I recorded an interview for TV Prima with reporter Laura Doubková. The topic currently filling headlines is simple and dramatic: because of artificial intelligence, chips are supposedly running out and electronics will become extremely expensive.

Is it true? Partly yes. But a headline like "chips are running out" is about as accurate as saying that the Czech Republic is running out of food because truffle oil has run out.

The report will appear on TV Prima in the following days. And since only a short clip can fit on TV, I'm giving a longer version of the whole idea here. This time also with a glossary of terms, because without them one can easily get lost in this debate.

What is actually happening

We experienced something similar between 2020 and 2022. Then the pandemic was the cause: closed factories, broken logistics and almost everything was missing, including simple chips that move the window up and down in the car.

Today's situation looks similar at first glance, but the cause is different. The world is building huge data centers for artificial intelligence. And they need the most valuable that the semiconductor industry can produce: the fastest memories, the most modern manufacturing processes, advanced packaging and top AI accelerators.

Data centers are an extremely lucrative customer for manufacturers. They are able to prepay for capacity years in advance and pay a higher margin. The ordinary customer with a cheap laptop or phone thus becomes secondary for the manufacturer.

An analogy that works well: imagine that there are only a few large bakeries in the world. And suddenly a very rich customer comes along and orders thousands of premium meals for years to come. Bakeries bake mainly for him, and there is less capacity left for ordinary customers. This is exactly what AI data centers are doing with the memory market and the most advanced chips.

An important nuance that headlines often miss: according to Deloitte, high-value AI chips account for roughly half the value of the entire semiconductor market, but less than 0.2 percent of unit volume. AI is not consuming all chips. It is mainly pulling the most expensive and rarest layers of production toward itself. Most classic chips made on older technologies are not the core problem.

All chips are missing. The most expensive, fastest and most modern ones are mainly missing.

A little vocabulary, let's speak the same language

A Wafer is a circular plate made of extremely pure silicon, on which hundreds of chips are produced at once. It is the basic "canvas" of the entire industry. With the most advanced generations, we are not talking about cheap raw materials, but about production capacity worth tens of thousands of dollars per board.

Nanometers such as 3nm, 5nm or 7nm indicate the generation of manufacturing technology. The smaller the number, the denser, more powerful and more economical the chip. And also the more expensive and scarce production capacity.

DRAM is operational memory. In computers and servers, we know it as DDR, in phones as more economical LPDDR. When they say they make memory more expensive, this is often what is meant.

NAND flash is memory for long-term data storage: SSD drives in computers, storage in phones and data storage in servers.

HBM, or High Bandwidth Memory, is the crown-jewel category. It is extremely fast memory stacked in layers and placed right next to an AI accelerator. It is expensive, difficult to produce, and consumed by AI in large quantities. HBM is one reason manufacturers are moving some capacity away from more common memory products.

And why is memory so crucial to AI?

The processor calculates, but the memory feeds it data. Without fast enough memory, even the most expensive AI chip is like a race car standing in a queue.

I can also see it in practice when I run large models. The first question is often not how fast the processor is, but how much memory the system has and how fast data can get into it.

Who are the main players?

TSMC is a Taiwanese contract manufacturer that produces chips for companies such as Apple, Nvidia, AMD or Qualcomm. Its quarterly results show nicely where the market is moving: high-performance computing and AI are the main growth drivers for TSMC today, while consumer electronics is losing relative weight.

Samsung and SK Hynix are among the largest memory manufacturers in the world. It is they who decide how much production capacity will go to HBM and server memories and how much will remain for ordinary laptops, phones and SSDs.

Micron is the third major memory player. Its announced investments in the United States show that the industry is taking the new demand seriously. But a high-end memory factory is not a hall that you open in half a year. It's years of building, tweaking yields, and training people.

Nvidia stands on the other side of the equation. Its AI accelerators are one of the main reasons for the huge demand for HBM and advanced encapsulation. It is not the only cause of the entire crisis, but it is its most visible symbol.

Why isn't it covid 2.0

The crisis of 2020 to 2022 was widespread. Older and simpler chips for cars, washing machines or industrial equipment were also missing because factories and ships were expensive. It was enough to restart logistics and production, and the problem gradually disappeared.

Today's crisis is structural and selective. It does not apply to all chips equally. It mainly concerns the most modern processes, advanced encapsulation and memories that AI needs in huge quantities. And it will not disappear when container ships start running, because the cause is not only a broken chain, but a conscious decision by producers to sell capacity where the margin is highest.

Paradoxically, it is less visible to the customer. There are no empty shelves in stores. But the price wave may last longer.

What does this mean for the Czech wallet?

This is no longer just an abstract forecast. TrendForce expects contract prices for conventional DRAM to increase by 13 to 18 percent and NAND by 10 to 15 percent quarter-on-quarter in the third quarter of 2026. For server DRAM, it reports growth of 13 to 18 percent for the same period.

Gartner estimates that a surge in memory prices will reduce global PC and smartphone shipments in 2026, while raising average device prices. Omdia adds an important social detail to this: the hardest hit may not be the luxury segment, but cheap phones under $400, where memory makes up a huge part of production costs.

This is crucial. When they make memory more expensive in an expensive phone, the manufacturer still has some margin to work with. There is almost no space for a cheap device. The result can be twofold: either a higher price or worse parameters. And sometimes both.

New products may have less memory than older generations. With the cheapest electronics, it can happen that the customer gets a newer model, but with worse long-term usability.

In addition, the increase in the price of components is prescribed to stores with a delay. First, the memory contract will become more expensive, then a new delivery to the manufacturer, then the finished device, and only finally the price tag in the e-shop. Therefore, the official statistics may show calmness for a while, while the pressure inside the supply chain is already growing.

How long will it take?

The honest answer: no one knows for sure.

New capacity is being built, but the semiconductor industry has a long history of inertia. Investments in the order of tens to hundreds of billions of dollars do not mean that it will be done in a few months. For memories and advanced chips, it is calculated in years.

On the other hand, there is a counterargument that I take seriously: AI investments may be overkill. If it turns out that data center returns aren't coming as quickly as investors expected, demand could falter. The memory market is historically cyclical and can turn around surprisingly quickly.

That's why it makes no sense to panic or stockpile electronics. Anyone who buys today "to be sure" can buy exactly at the price peak.

Circular economy as the smartest answer

This is the main positive message of the whole topic for me. The mentality of high-speed electronics - buy, use for two years, throw away - ceases to make sense economically and ecologically.

Rather than running to the store and pre-stocking, I think it's smarter to think about getting something that will last a long time. Typically a refurbishable device where you can add memory, replace the battery, clean the cooling and extend the life by years.

Practically this means:

• When buying a new device, pay attention to whether it has soldered memory or whether it can be expanded later.
• For an existing computer, consider upgrading instead of replacing: RAM, SSD, battery, cooling cleaning.
• For companies, think about a longer life cycle of technology, not just about a general change according to the calendar.
• Take refurbished technology as a full-fledged alternative, not as an emergency solution.

A professionally refurbished executive notebook can offer excellent quality at a fraction of the price and is not as sensitive to current semiconductor inflation. At a time when new cheap devices can save on memory and storage, this makes even more sense than before.

Most importantly, don't panic

I would formulate the golden rule simply:

Buy according to real need, not according to headlines. Whoever needs the device should follow the prices of the current stock and not hesitate indefinitely. Anyone who has a functional device should first consider whether it can be expanded, repaired or let it serve longer.

The AI revolution has real physical costs. It lives not only in the cloud, but in factories, wafers, memory lines, electricity consumption and electronics prices. We will all feel some of these costs now.

But the answer is not hysteria or buyouts. The answer is to behave rationally, think circularly, and not be swayed by either the catastrophists or those who trivialize the problem.

Sources and Links

• Čipy docházejí. AI vyžene ceny notebooků a mobilů do extrémů - Czech article that opened the topic to a wider public.
• Deloitte: 2026 Global Semiconductor Industry Outlook - context on value concentration in AI chips.
• TrendForce: AI Server Demand Continues to Support Memory Prices in 3Q26 - DRAM and NAND price outlook.
• TrendForce: Server DRAM Contract Prices Expected to Rise 13-18% QoQ in 3Q26 - server memory detail.
• Gartner: Surging Memory Costs Will Reduce Global PC and Smartphone Shipments in 2026 - expected impact on PCs and smartphones.
• TSMC: 2026 Q1 Quarterly Results - financial results and demand shift signal.
• Micron: Accelerates U.S. Investments - new memory manufacturing investments.
• Omdia: Global smartphones priced below $400 will decline by 22% as memory costs soar - impact on affordable phones.