SEOUL, August 08 (AJP) - Ask two leading artificial-intelligence chatbots why SK hynix has lost more than half its value from its June peak, and they arrive at a remarkably similar conclusion: the company's business has not collapsed, but the price investors were willing to pay for its extraordinary AI-driven earnings has.
AJP separately asked Anthropic's Claude and OpenAI's ChatGPT to examine SK hynix's rout in Seoul and New York, including trading flows following its Nasdaq debut, the reasons behind the selloff, analyst views and the upside and downside risks ahead.
Both identified essentially the same fault lines — excessive positioning and leverage, an earnings bar that had become almost impossible to clear, worries about the sustainability of AI spending, rising Chinese competition and investor frustration that record cash generation has not yet translated into equally extraordinary shareholder returns.
Where they differ is more revealing.
Claude puts greater weight on the Nasdaq listing itself as a "sell-the-news" event that encouraged investors to sell the Korean shares while buying the scarcer U.S. receipts.
ChatGPT sees the ADR less as the cause than as an amplifier: SK hynix was already retreating from its June peak before the July 10 listing, but the creation of a second market gave global investors another venue through which sentiment, leverage and overnight price discovery could ricochet back into Seoul.
Both interpretations can be true.
SK hynix closed at a record 2.919 million won on June 22, when it overtook Samsung Electronics to become South Korea's largest listed company for the first time in more than a quarter century. It subsequently touched an intraday record of 2.987 million won on June 25 before closing at 2.917 million won. Friday's close of 1.422 million won leaves the stock more than 52 percent below that intraday peak.
That means the decline had already begun before the ADRs arrived.
Still, the Nasdaq debut fundamentally changed SK hynix's market structure.
The company sold 177.9 million American depositary shares at $149 apiece, with every 10 ADRs representing one Korean ordinary share, raising $26.5 billion through 17.79 million newly issued underlying shares. The offering therefore added shares equivalent to about 2.5 percent of the existing stock.
Demand was enormous. But rather than immediately pulling the Korean stock upward toward a higher U.S. valuation, the listing created two prices for essentially the same company.
Within days, the ADRs traded at a premium of about 36 percent to the Seoul shares. U.S. investors could cancel ADRs and receive Korean stock, but creating new ADRs out of existing Korean shares remained slower and more restricted, limiting the arbitrage that would normally close such a gap.
This is where Claude's analysis is strongest.
It focuses heavily on flows. Foreign investors were selling SK hynix in Seoul as U.S. demand poured into SKHY, securities lending increased ahead of the listing and Korean investors themselves began buying the U.S. security.
In that reading, Nasdaq did not simply broaden the shareholder base. It temporarily diverted marginal demand away from Seoul and offered traders a new relative-value trade between the expensive ADR and cheaper ordinary stock.
ChatGPT places more emphasis on transmission. Once SKHY began trading in New York, SK hynix effectively became a nearly round-the-clock stock. A sharp ADR move after Seoul closed could become the reference point for the next Korean session, while movements in the KOSPI fed back into New York hours later.
The distinction matters because it produces different conclusions about the ADR premium.
Claude views the premium chiefly as an overhang: as conversion capacity and market efficiency improve, the expensive U.S. shares could fall toward Seoul parity.
ChatGPT sees the same premium as partly structural. Restricted fungibility, strong U.S. institutional demand and limited ADR supply could allow a scarcity premium to persist, much as it has for Taiwan's TSMC. Reuters Breakingviews similarly argued that restrictions on creating new SK hynix ADRs could keep the U.S. premium alive.
What both AIs agree on more firmly is that the ADR cannot explain a 50 percent collapse by itself.
The bigger problem is what happened to expectations.
SK hynix delivered operating profit of 60.5 trillion won in the second quarter, more than six times its year-earlier result and an all-time record. Yet analysts had expected about 64 trillion won. Revenue of 79.3 trillion won also undershot the roughly 84 trillion-won consensus. The shares fell 9.6 percent after the earnings release.
A company reporting record profits was punished because the market had priced in something even better.
Both Claude and ChatGPT identify that as a turning point. The issue ceased to be whether SK hynix was making enormous money and became whether its extraordinary margins represented a sustainable new structure for memory chips or the most profitable point of another cycle.
There are reasons for doubt.
SK hynix's dominance in high-bandwidth memory ironically left it less exposed than Samsung to some of the steepest increases in conventional DRAM prices. Long-term HBM supply contracts provide greater demand visibility but can also limit how quickly prices can be reset upward. Delayed shipments of HBM4 contributed to the second-quarter miss.
Then came broader questions about AI itself.
Investors have begun asking whether hyperscalers can continue funding hundreds of billions of dollars in data-center construction at the same pace, particularly as scrutiny grows over the returns generated by the investment. China simultaneously supplied another source of anxiety through CXMT's expansion and reports of progress in domestic lithography equipment, raising the possibility that Chinese memory producers could eventually add enough supply to undermine the oligopoly enjoyed by Samsung, SK hynix and Micron.
The market mechanics made every concern more violent.
JPMorgan estimated that assets in leveraged ETFs tied to Samsung Electronics and SK hynix fell from about $50 billion in late June to $17 billion by late July. Some global investors consequently described Korea's crash as a "leverage event" rather than an earnings event, while JPMorgan estimated hedge-fund deleveraging was largely complete by early August.
That observation is central to the bullish case presented by both AI systems.
If forced selling rather than collapsing earnings produced much of the decline, removing leverage can produce just as powerful a rebound. SK hynix demonstrated that on July 31 when it hit its 30 percent daily upper limit amid record foreign buying, only to resume falling days later.
But Claude gives somewhat more weight than ChatGPT to the possibility that the memory narrative itself is changing. It highlights weakening NAND signals, questions around future HBM specifications, Chinese capacity and the possibility that hyperscalers eventually digest the massive infrastructure they have already ordered.
ChatGPT puts more emphasis on the contradiction between the share-price collapse and what customers are actually telling SK hynix.
Management said after its second-quarter results that major customers were continuing to ask for additional memory supply. SK hynix is negotiating five-year supply agreements and raised its 2026 capital-spending plan into the high-40-trillion-won range. That is not how a memory producer normally behaves when it sees an imminent demand collapse.
Friday's 54.3 trillion-won investment decision sharpens that debate.
The board approved 35.2 trillion won for the second Yongin fab and 19.1 trillion won for the M17 NAND plant in Cheongju. But neither represents immediate supply. The first cleanroom at Yongin Y2 is not scheduled until June 2029, while M17's is expected in December 2028.
That timing also corrects one easy interpretation of Friday's market action. The Reuters report carrying the capex announcement was published at 8:17 a.m. UTC, or 5:17 p.m. in Seoul, after the Korean market had closed. The announcement therefore cannot explain SK hynix's 4.88 percent decline during Friday's regular KOSPI session.
It can, however, reinforce the question investors are increasingly asking: what will SK hynix do with all the cash?
Net cash had reached 88 trillion won at the end of June, according to the company, while investors have become increasingly vocal about demanding dividends and buybacks.
SK hynix and Samsung currently target shareholder returns equal to about half of free cash flow, compared with Micron's commitment to return 100 percent. Janus Henderson portfolio manager Richard Clode has argued SK hynix should lift that figure to at least 80 percent.
That concern produces perhaps the clearest point of agreement between Claude and ChatGPT.
The next major catalyst may not be another record profit. It may be capital allocation.
SK hynix said Friday it would announce additional shareholder-return measures during the third quarter, alongside a quarterly dividend of 375 won per share.
A large buyback and cancellation program would answer the market's concern that management intends to accumulate cash and pour most of it back into fabrication capacity at the top of the cycle. A modest program would instead reinforce the view that management itself is unwilling to treat today's earnings as permanent.
The analyst divide reflects exactly that uncertainty.
The bullish camp argues the fundamental AI-memory shortage remains intact and the stock has fallen far more than earnings estimates justify. Six U.S. brokerages began covering SK hynix's ADRs with bullish ratings in early August, with Rosenblatt setting a $320 target and Bank of America emphasizing the company's leadership in high-end memory.
At the other extreme, BNK Investment & Securities cut its Seoul target to 1.48 million won and maintained a Hold rating, arguing that slowing demand momentum and continued industry investment would limit the rebound.
The enormous spread itself is informative.
Analysts are no longer arguing primarily about this quarter's earnings. They are arguing about which earnings multiple SK hynix deserves once today's shortage ends — and when that ending comes.
Combining the two AI analyses produces a relatively simple map of what happens next.
The downside case requires several things to start going wrong together: HBM pricing power shifts toward Nvidia and other customers; NAND weakens first; hyperscalers slow infrastructure investment; CXMT accelerates capacity; SK hynix continues spending aggressively while delivering disappointing shareholder returns; and the ADR premium contracts through a fall in SKHY rather than a rise in Seoul.
The upside case is almost the mirror image. HBM4 shipments accelerate in the second half, multi-year contracts demonstrate that demand is more durable than in past memory cycles, shortages persist because new fabs cannot arrive quickly enough, leveraged selling exhausts itself, and SK hynix uses its cash mountain for a sufficiently large buyback or cancellation to force investors to rethink the Korea discount.
There is one important limitation to calling this an "AI consensus."
Claude and ChatGPT are not two independent portfolio managers making original forecasts. Both synthesize many of the same analyst reports, company disclosures and financial-news stories. Agreement between them therefore does not amount to two independent investment votes.
Still both end up at essentially the same near-term test: shareholder returns, HBM4 execution and the durability of AI capital spending will matter more for SK hynix's next major move than another headline showing record quarterly profit.
Put together, the AI verdict is neither comfortably bullish nor conventionally bearish.
SK hynix has already lost the valuation of an AI company. The question now is whether it also loses the earnings of one.
Copyright ⓒ Aju Press All rights reserved.