Up 330% in 6 months and still cheap. Armina Rosenberg on her top growth pick for 2026

For those uninitiated in the world of high bandwidth memory, South Korean chipmaker SK Hynix (KRX: 000660) probably wasn’t a familiar name just a week ago.

However, anyone who was paying attention to Minotaur Capital co-founder and portfolio manager Arms Rosenberg when she gave Livewire her top growth stock pick for 2026 would have already had SK Hynix on their radar.

“You’re seeing memory emerge as a clear beneficiary of this whole AI supercycle, and I think we’re still at the early innings of that. So, even though DRAM prices are up something like 150% since their 2023 trough, and some components are actually up, I think three to four times over the last couple of months, we still think that the beneficiaries of that are going to be the three big players. So there's Micron, Samsung, and SK Hynix. We like SK Hynix the most out of that,” Rosenberg said in December.

Since the video was recorded, SK Hynix is up about 330%.

Over the same time period, the other players she mentioned - Micron (NASDAQ: MU) and Samsung (KRX: 005930) - are also up 280% and 210%, respectively.

Is $1 trillion fair value?

Following the recent boom in its price, Rosenberg says she is “obviously feeling pretty good about it”.

“We potentially weren't in March when it was coming back, but we saw no fundamental reason for the pullback in March. We thought it was a bit of a scare that was not dissimilar to the scare we saw for Nvidia with DeepSeek,” she says.

“Google had put out a paper on Turbo Quant, effectively saying that memory prices were going to compress, so we were still positive after that and we used that drawdown in March to add to the position in both SK Hynix and Micron, which are our two biggest positions. They have been almost since February at least.”

Minotaur has since “trimmed around the edges” on its exposure to the chipmakers, but not out of any concern with the companies themselves, but the concentration within its portfolio.

“I think AI infrastructure as a whole bucket for us got quite large, and we are very mindful of risks in our portfolio and from a thematic perspective, so we've trimmed around the edges, but they're still our two biggest positions<” Rosenberg explains.

“We still believe in it fundamentally and we see memory being the real hidden choke point in the AI build out, because every AI server needs them, you can't run a large language model without enough HBM, so that's a key driver of the sort of AI infrastructure dynamic.”

Thomas Rice, also a co-founder of Minotaur, adds that despite the rapid growth of SK Hynix, it’s still undervalued.

“We think there's less upside than there was a month ago, but we don't look at the movement in the last few months and think this is a huge bubble. I think it’s reflecting the reality of growth in AI and how important memory is, and also the shifting dynamics in the industry,” Rice says.

Even more bullish is Rosenberg, arguing that SK Hynix is the “most important company in the AI chip supply chain after Nvidia itself”.

“Without SK Hynix’s HBM, the Nvidia GPUs don't work, and it still trades well less than the multiple of Micron,” she says.

This is in part due to its South Korean listing, which limits the ability of foreign investors pouring cash into the AI boom to access SK Hynix directly. A US listing, which Rosenberg says the company “keeps threatening”, could boost its price even further.

“It's around 2 million Korean won at the moment, I think our bull case is at around 3 million or above,” she adds.

“That's still around 40% upside from here.”

What’s driving the AI chipmaker trade?

Warnings from the Pope aside, there is very little in the way of headwinds for companies in the AI space.

It’s gotten to the point where a Japanese toilet manufacturer pivoting to AI can reach a five-year high.

The answer to why semiconductors are booming right now is down to the two main components of their manufacturing: development and design.

Global X investment strategist Billy Leung explains that while Nvidia essentially owns the semiconductor design space, how they actually construct the chips in a way that is efficient and doable is the key.

“This is where Samsung, SK Hynix, and to an extent even Micron comes in,” Leung says.

“The latest craze with SK Hynix is because there is this new type of memory called high bandwidth memory. It’s the same thing as the normal memory chips, but they're just stacking it differently, which means that now it’s much more efficient and that same size chip can actually do more than before.

“The latest breakthrough is that generally SK Hynix has been increasing and improving their technology, their ability to produce better chips, and this is what's really rerating the whole story.”

Also bolstering the South Korean chipmakers is what Leung calls the liberalisation of the Korean market.

“The second part of the story is more about Korean regulators,” he explains.

“About two or three years ago they did the same thing that Japan regulators did to the Japanese equity market. They told these companies, ‘look, we've been very much focused on the family oriented business model, whereas now can you consider returning more wealth to shareholders, improving your corporate governance, improving the core value of your company’.

“That was actually a big step for the improvement of valuation and the efficiency of these companies.”

The memory bottleneck

At the same time, there is a massive squeeze on memory for AI, with the demand far outstripping the supply and it taking years to increase manufacturing capacity.

The move from AI training to AI inference has been a key factor in creating this “huge bottleneck”, according to Leung.

“AI training is when you train a machine to learn what an Apple is. AI inference is when you actually bombard the machine with a lot of apples and tell them to identify it,” he says.

However, this is also sparking growth for AI companies and has dragged forward a lot of the demand for semiconductors.

For the manufacturers, this has driven expected earnings even higher.

“Valuation actually hasn’t caught up with the expectations of growth, so I would say that the valuation is still somewhat sustainable from a valuation to growth perspective - or what I call bank for buck,” Leung says.

“Obviously we’re going to see further explosion of AI usage. It takes about three to five years to actually build a semiconductor plant, then it takes probably another one to two years to actually get up and running into that quality and capacity.

“It takes a long lead time to actually ramp up semiconductor capacity, which also means that this bottleneck will last longer, and hence even though we're seeing higher valuation it’s very strongly supported by tail winds.”

Rosenberg also backed the memory crunch to be a long-term issue driving up valuations for the likes of SK Hynix.

“There is a dynamic where in the past these memory companies have been a bit more cyclical, more valued on a commodity cycle basis, and that's because they do tend to see a lot of demand and then build for that demand, then be over capacity for a little while, and then there was a bit of a catch up for that demand to meet supply,” she says.

“What we’re seeing with this demand for AI infrastructure, we think the tailwinds are a lot more structural this time around, and you're not really seeing capacity come online until around 2028.”

It might be hard to believe that a stock that is up 330% in six months is "still pretty reasonably priced", as Rosenberg puts it, but the memory bottleneck makes a compelling case.