By Noah Weidner

Investors are possessed by a fear of AI and fears from AI. It’s hard to know which fear is stronger, but both are already shaping the market. Depending on where you’re sitting, it could be an opportunity.

On one hand, you have investors worried about the impact that fast-progressing AI models by Anthropic, Google, and OpenAI will have on the market — year-to-date, the SPDR S&P Software & Services ETF ($XSW) has fallen more than 17.9%. Another software-centric ETF, the iShares Expanded Tech-Software Sector ETF ($IGV), has done one worse; it’s down 20.1%.

Those declines are stark, but they’re not singular. In fact, investors aren’t too sure about Magnificent Seven tech stocks, either. Since the start of the year, AI first-movers like Microsoft (-15% YTD), Amazon (-9.6%), and Tesla (-6.13%) have been punished as investors have digested the enormous costs associated with their respective AI buildouts.

So, what do you do? Do you buy the software dip, betting that the smart money on Wall Street is wrong about the disruption? Or, alternatively, do you buy the Big Tech dip, betting that Wall Street is wrong about the payoff? At this point, perhaps the answer is neither, at least not before you consider this hedge on both.

What Wall Street has wrong about AI

Last week, I made the case in our TheStreet Daily newsletter that there are plenty of industries that AI could eventually disrupt. However, many of the smart people on Wall Street are completely blind to the industries that could interrupt the AI rollout.

Big Tech has committed an unmistakably large fortune to building frontier AI models, acquiring compute, and placing it in data centers. Still, there are huge bottlenecks. But here, the well-to-do on Wall Street have made a few leaps of judgment. That is to say, they have skipped to the conclusion that the AI has already won.

There is precedent for this. Some recent research from Goldman Sachs showed that investors had largely checked out of U.S. newspaper stocks before firms’ consensus forward earnings fell off a cliff. Newspaper stocks had long before begun a decade-long downward spiral—even before the impacts to print were palpable. Let that serve as evidence that the market sometimes demonstrates foresight.

But here, the market has skipped over a few massive developments. The first is the actual application of the technology. In other words, how will businesses use them? Many firms are still working that out, exploring the possibilities.

Another assumption is that there are no constraints on AI implementation in the workplace. Anybody who has ever worked in a corporate environment knows that is simply not true. Advocacy, implementation, and proving takes time. These matters are only complicated by internal politicking over what work the AI will do or eventually replace.

But chief among these factors is the cost. You don’t generally see it talked about; you just see the circular movement of capital. At this stage, AI models sort of remind me of Uber. Only, Uber never lost this much money trying to win over consumers from more expensive, regulated taxicab services. The hope is that, in the end, you have shifted customer behavior so dramatically that people replace your predecessor with your name. They turn your app into a verb. Nobody “hails a taxi” anymore; they “get an Uber.” That’s why the company was able to pull the levers and become a profitable firm.

What does AI have to do with Uber? Same general concept of losing money, only your competition is human capital. What AI models are offering is a productivity booster; an excuse to say the word “efficiency” a few hundred times on your earnings call and dropkick 10,000 employees from your payroll. What Wall Street, and arguably management, does not understand is that the ability to win in this market is largely dependent on unit economics.

At this stage, we know that virtually every query made with AI is a loss-leader. Companies will try their best to spin how they make money off of their frontier AI models, but they lose money every time you ask a question. There’s all this inventive accounting taking place to try to make sense of the more than $1 trillion in capital expenditures being pumped into this technology, but make no mistake: Somebody is going to have to pay for all of this.

There is this cope that you hear from AI optimists that we’re going to be living in some post-scarcity utopia in the next five years; that you have little time left to “escape the permanent underclass.” From this vantage, I don’t see how OpenAI CEO Sam Altman could be right about “the costs of intelligence and energy” trending on a path toward near-zero. That is a decades-long commitment.

For those reasons alone, I wouldn’t rule AI’s potential out, but I wouldn’t get too attached either. From this juncture, it’s obvious that the industry is going to take on water, and somebody is going to fold in all this spending. However, the long tail generally benefits the technology. Smart investors have already arbitraged a lot of the opportunity here by buying up electricity generation plays, AI chipmakers, and even data center lessors.

However, they’ve not closed the door on at least one outstanding opportunity.

South Korea, my beloved

In the AI idealist’s most perfect world, there will eventually be enough energy, hardware, and equipment to run the models affordably. But at least at this stage, there are massive limitations; these are only exacerbated by competitors vying for the limited resources. They are treating this situation as a “winner-take-all.”

Wall Street has already concluded that those who sell pickaxes are likely to make more money than gold miners. They have, however, mispriced one of the bigger pickaxe sellers. Samsung and SK Hynix, two of South Korea’s most valuable companies, are among the only three companies producing AI-capable flash memory at scale.

READ: Samsung co-CEO warns that ‘unprecedented’ memory shortage could impact smartphone, laptop costs

Investors have already done a number on the third viable flash producer in U.S-based Micron Technology, pushing its stock up more than 33.5% year-to-date amid jitters over the flash memory storage. But despite boasting a more premium valuation relative to its Korean competition, Micron lags both SK Hynix and Samsung in production.

Put plainly, this is a huge constraint that cannot be patched over quickly. Sure, Chinese players could enter the high-bandwidth memory market, but that would take years. The fervor in high bandwidth memory is a problem now, which is impacting the price of all compute equipment. DDR5 RAM is turning into rare loot as AI sucks up the available supply from other industries and use cases.

Korea discount at play

One factor that can explain the difference in valuations between the U.S.-based Micron and Korea’s Samsung and SK Hynix is the so-called Korea Discount. It helps to explain why a country so instrumental in the artificial intelligence (AI) boom is being appraised at a discount.

In short, the Korea Discount has a lot to do with the control of domestic corporations. Conglomerates in the 13th largest global economy are generally controlled by family “wealthy cliques” called chaebols. They are said to exert massive influence over corporate governance and management, as well as keeping share prices low.

However, the dominance of the chaebol has been slowly breaking, in part because the country understands that the discount is weighing on economic growth. However, a bigger factor is the sheer amount of outside capital that is now funneling into South Korea. That has helped push the companies’ P/E ratios out of the tepid territory they inhabited for over 20 years.

Still, despite the renewed interest in the South Korean economy, the stock market remains at a discount relative to U.S. equities. The case isn’t that the Korea Discount is going to completely disappear, but at current valuations, it still has the potential to be an attractive opportunity for those who believe the AI gravy train will continue.

What’s the buy?

Buying stock directly in SK Hynix or Samsung probably won’t be super practical for most retail traders. A healthy alternative is just buying an exchange-traded fund (ETF) made up of South Korean stocks, which should offer enough exposure to those companies, along with a healthy foundation of firms that are not as exposed to the memory trade.

The Franklin FTSE South Korea ETF ($FKLR) is the cheapest ETF that I could find, which tracks the South Korean markets, with over 41% of its holdings in Samsung Electronics and SK Hynix. At a net expense ratio of just 0.09%, it’s almost as cheap as a developed markets ETF from your favorite ETF provider. It’s also up more than 120% over the past year.

Another option is the more expensive iShares MSCI South Korea ETF ($EWY), which counts 47.3% of its holdings in the two compute giants. With a bias towards the country’s large and mid caps, this ETF has done one better on $FLKR, up 131% over the past year, maybe justifying its higher 0.59% expense ratio.

This article was originally published by Thestreet