Who is shouting "bubble"? And why memory, of all things?

Micron is currently priced at around 8 times next year's expected earnings. Eight. That is not a typo. For comparison, the semiconductor sector typically trades at around 37 times, the Nasdaq-100 at around 25-28 times, and Nvidia at around 30 times. If you look at growth-adjusted pricing (PEG), Micron is down around 0.06 – which roughly means you are paying a fraction of what the earnings growth would suggest.

There are many words you can use to describe such pricing. "Bubble" is not one of them.

The selective bubble vocabulary

There is something worth reflecting on here: Who shouted "bubble" when oil companies rose 200 percent from 2020 to 2022 in line with oil prices? Who is shouting "bubble" at defense stocks that have doubled and tripled since 2022? Who shouted "bubble" at Moderna and BioNTech in 2020 and 2021, when the whole world suddenly needed vaccines?

The answer is: quite a few, and certainly not with the same intensity.

Take Kongsberg Gruppen, which is perhaps the most striking example. The stock has risen over 1,500 percent in five years. It is priced at around 35 times next year's expected earnings. The trailing P/E is over 40. The market value has surpassed several hundred billion kroner. Yet, I haven't seen many people shouting "bubble" at Kongsberg. On the contrary, the consensus is that the company is "delivering on fundamentals," that NATO budgets provide "structural demand," and that the order book provides "multi-year visibility."

All of this is true. But it is at least as true for Micron – with the small difference that Micron is priced at a quarter of that. If Kongsberg at 35 times expected earnings is a rational defense rally, what is Micron at 8? The price performance over the last five years is also quite similar.

The difference does not lie in the fundamentals. Oil companies are at least as cyclical as memory manufacturers – probably more so. Defense stocks are driven by a geopolitical cycle that can turn faster than people think. Vaccine stocks were, by definition, dependent on a specific, time-limited event. Yet these rallies were mainly described as sensible reactions to changing fundamentals.

When meme stocks do exactly the same thing—reacting to a dramatic change in fundamentals—the cries of a bubble emerge as if they were rehearsed in advance.

What the market is actually pricing in

First, a few basics for those who don't follow the semiconductor market daily: Memory chips are roughly divided into two types. DRAM is the working memory found in everything from your mobile phone to the servers in Google's data centers—short-term storage that the processor retrieves data from constantly. NANDis permanent storage, meaning what is found in SSDs and memory cards. And then we have a new third category that has exploded with AI: HBM (High Bandwidth Memory), a specialized variant of DRAM where 8 or 12 chips are stacked on top of each other to provide enormous bandwidth. It is HBM that feeds Nvidia's AI accelerators, and it is HBM that only three companies in the entire world are capable of making—Samsung, SK Hynix, and Micron.

With that in place: What is driving the memory market now is not a typical cycle. It is something else, and it is worth being precise about what:

HBM is sold out – and contracts are getting longer.Micron has confirmed in its own quarterly presentation that its entire HBM production for 2026 is sold out. The same applies to SK Hynix and Samsung. What is even more interesting is the nature of the contracts: Customers are now signing agreements that extend 3–5 years into the future—a structural break from the industry's historical quarterly negotiations.

Production has been redirected. The big three have moved over 90 percent of their capacity to HBM and server-class DRAM. One HBM wafer (a silicon plate from which chips are made) displaces two or more wafers of standard DRAM because the chips are larger and the packaging is more complex. This means that the total number of memory chips in the world is actually going down, even though the number of wafers remains the same.

Demand is structural, not cyclical.AI training, AI inference, AI agents running around the clock, robots, drones, self-driving cars—all require absurd amounts of memory. Bloomberg Intelligence estimates that AI servers alone will account for over 60 percent of global DRAM consumption by 2030, up from 32 percent in 2020.

Supply cannot respond quickly. A new factory costs at least 15 billion dollars and takes 18 months or more to build. IDC estimates that DRAM supply will grow by only 16 percent in 2026—well below the historical norm of 20–30 percent, and far below the rate at which demand is growing.

Prices reflect this. DRAM prices are up 80-90 percent in a single quarter. TrendForce expects DRAM to become over 70 percent more expensive throughout 2026 as a whole. Bank of America expects industry DRAM revenue to grow 51 percent this year.

This is not a narrative. These are signed contracts, actual capacity decisions, and market prices.

When Elon Musk says "we'll buy everything they make"

Few signals regarding what is happening on the demand side are more honest than this. When Musk launched Terafab earlier this year, he said this verbatim about his existing suppliers:

"We're very grateful to our existing supply chain – to Samsung, TSMC, Micron, and others. And we would like them to expand as quickly as they can. And we will buy all of their chips – I've said these exact words to them. But there's a maximum rate at which they're comfortable expanding, and that rate is much less than we would like. So we either build the Terafab or we don't have the chips."

The figures he presented are almost cartoonishly large: Current global chip production is around 20 gigawatts per year. What Musk believes SpaceX, xAI, and Tesla alone will need annually is 1 terawatt– that is, 50 times the entire world's current production, from a single buyer.

You don't have to believe in Musk's ambitions to see the point. The important thing is what he is actually saying: The three dominant memory manufacturers cannot expand fast enough for one of their biggest customers, and this customer must therefore build their own factory – not to take the market from them, but because they cannot deliver enough. When one of the world's largest buyers publicly declares that "we'll buy everything they make," that is not a bubble signal. It is the exact opposite.

The defense parallel no one is drawing

Back to Kongsberg for a moment, because there is a connection here that people are overlooking.

The defense industry is being treated as a rational repricing because we all see what is driving it: war in Europe, rearmament, NATO targets of 2-5 percent of GDP, and order books that extend a decade into the future. The logic is transparent and easy to explain.

But What is actually driving the growing portion of defense budgets? More steel and more soldiers? To a certain extent. But what is growing the fastest is something quite different: drones, autonomous systems, missile defense, surveillance platforms, unmanned vessels, AI-driven target recognition. Exactly what Kongsberg Discovery and Kongsberg Defence & Aerospace deliver.

All these systems have one thing in common: They need memory. A lot of it. An autonomous drone is literally a flying computer with sensors. A self-driving truck is the same thing on wheels. A missile defense system that needs to track hundreds of simultaneous targets in real time requires HBM in the same way as an AI training cluster.

The defense supercycle and the memory supercycle are not two different topics. They are the same topic, viewed from two angles. One angle is treated as rational and priced at 35 times expected earnings. The other is called a bubble and priced at 8. There is no logically consistent way to hold both views simultaneously.

Why can the commodity oil go to "infinity," but not memory?

This is the core question, and I have yet to see anyone answer it properly.

Oil is a commodity. When demand exceeds supply, the price goes up – often violently – and the earnings of oil companies explode. No one questions the underlying logic. We understand that a shortage of oil is a shortage of oil.

Memory is also a commodity. DRAM and NAND are priced on spot and contract markets just like oil. The supply is controlled by three companies – Samsung, SK Hynix, and Micron together hold over 95 percent of the DRAM market, which is more concentrated than OPEC. Demand is driven by a structural consumption trend (AI) of the same type that increased energy consumption has been historically.

Yet: When the oil price triples, people buy Equinor. When the memory price doubles, people cry bubble at Micron.

The only consistent explanation I can find is about familiarity. We have understood oil for a hundred years. Oil is physical, it comes out of the ground, it is something we know. Very few have heard of HBM stacks with 12 layers of silicon feeding Nvidia's GPUs. It feels like "tech hype" even when, in reality, it is industry.

But the market doesn't care about what feels familiar. It cares about cash flows.

The connection that is a little too obvious

A small observation: There is a strikingly strong correlation between those who are heavily long on oil and commodities, and those who are calling a bubble in memory. People who otherwise love to explain why scarcity of an input factor drives up prices and margins.

Two possible explanations:

1. They have conducted a thorough analysis and concluded that memory is fundamentally different from oil, in a way that justifies the opposite conclusion.
2. They are not participating in the rally and need a narrative.

In my experience, explanation number two covers more of reality than we like to admit. It is uncomfortable to sit on the sidelines while others make money. It is even more uncomfortable if you have a public profile as a trading vlogger or "value investor." Then you have to find a reason why the others are wrong – otherwise, you have to find a reason why you yourself did it.

"It's a bubble" is the emotionally cheapest explanation there is. It requires no humility. It lets you be the enlightened one. And if you happen to be right sometime in the future – no matter when – you can point back to your tweet.

The Nvidia parallel

Remember Nvidia before the split in June 2024. The stock had gone straight up, peaking at around 1200 dollars just before the company executed a 10-for-1 split. People were shouting bubble on Twitter every single day. The arguments were the same as those we hear about Micron now: cyclical, hype, AI is overrated, all the big buyers are four companies, this has to crash.

Today, Nvidia trades at around 220 dollarsAdjusted for the split, that’s around $2,200 on a pre-split basis—nearly double the level people were calling a "bubble" in 2024. The market cap is heading toward $5.5 trillion.

Why did this happen? Because earnings grew even faster than the stock, meaning the multiples actually went down during the rally.

And, not least: a continuous bid from passive management.

Micron has the same characteristics today: the stock is rising, but estimates for next year's earnings are rising faster—so the valuation relative to earnings is falling even as the share price climbs. That is what a structural repricing looks like. That is not what a bubble looks like.

The conclusion

You don't have to buy Micron. You also don't have to buy Samsung or SK Hynix. There are plenty of good reasons to sit on your hands.

But if you are sitting on the sidelines, do everyone a favor: don't call it a bubble. Call it what it is—that you didn't get in. That’s fine. There’s nothing to be ashamed of. We all miss trains.

What isn't fine is rationalizing it with an analysis that doesn't hold up. A stock at 8 times next year's earnings, in a market controlled by three companies, with production sold out on multi-year contracts, with Elon Musk publicly promising to buy everything they can produce, driven by the same structural wall of demand as the defense and drone market you’re happy to pay 35 times for—that is not a bubble.

It might be the beginning of a completely new valuation of what memory is actually worth in a world with near-infinite demand and limited supply.

Avanto writes about markets the way we like to read about them ourselves—without blinders, without the pressure of consensus, and with a bit more honesty about why people actually believe what they believe.

‍