One CEO's "Casino" Remark Took More Than 5% Off the Kospi. The Real Story Is Korea's Concentration in Two Memory Stocks

On Tuesday 13 October 2026 South Korea's Kospi fell more than 5%, with Samsung Electronics down about 7% and SK Hynix down about 10%. The trigger most coverage pointed to was a remark by Broadcom chief executive Hock Tan at an industry event in Singapore. According to the Options Trading Report, Tan said “a lot of investors have no clue what they are investing in,” that they held “very extreme expectations,” and that AI investing was “almost like they are in a casino.” Broadcom sells networking chips and custom AI processors, not memory. The more useful question than what Tan meant is why a sentence from a supplier moves two Korean memory makers by seven and ten percent, and why the same pair fell just as hard in June.

What was said and what moved

The sources agree on the substance and differ on detail. The Options Trading Report dates the remarks to Monday 12 October; a KuCoin news flash citing Bloomberg dates them to Tuesday 13 October. We write “this week.” On the index, CoinDesk reported a fall of 5.5% and the Options Trading Report said more than 5%. Other outlets printed larger figures, one of which appears to match Samsung's move rather than the index, so we use the lower, better-corroborated range and attribute CoinDesk's 5.5%. For single stocks, the Options Trading Report gave Samsung down 7.4% and SK Hynix down 10.1%; CoinDesk rounded to 7% and 10%.

The KuCoin flash, which we could not independently confirm, listed other moves attributed to Bloomberg: Japan's Nikkei down 1.8%, SoftBank down 8.5%, Tokyo Electron down 7.15% and Micron down 3.9%, with the Nasdaq Composite down 0.4% the previous day. The pattern is consistent across venues: the damage concentrated in memory and AI-exposed hardware, while broad indices fell far less. Korea took the largest hit.

Why Korea took the hit

Here is arithmetic the coverage did not do. The Kospi fell 5.5% while the two stocks fell about 7.4% and 10.1%. If every other stock in the index had been flat, Samsung and SK Hynix together would have to make up somewhere between 55% (5.5 divided by 10) and 74% (5.5 divided by 7.4) of the index to produce that move on their own. The sources do not give the two firms' index weights, and neither bound is plausible as a real weight, so the sensible conclusion is that the rest of the market fell too. That is the signature of an index where a large share of value sits in two names, so that their fall drags everything else with it through index funds, derivatives hedging and sentiment. The stocks are not the whole Kospi; they are enough of it that the Kospi cannot be separated from them.

Why would a remark about AI investors hit memory makers in particular? The common thread is that Samsung and SK Hynix are priced on expectations of AI-driven demand for high-bandwidth memory, which makes them a leveraged way to hold the AI trade. A statement that expectations are extreme is, to a trader, a statement about the multiple on those stocks. The Options Trading Report adds that the selloff hit a market with thin marginal buying, meaning few buyers were willing to step in at the prior price. Thin buying turns a sentiment shock into a price move.

Positioning appears to have amplified it. CoinDesk's derivatives wrap, which tracks futures data, reported that SK Hynix open interest rose about 18% while its price fell about 8%, which it calculated as roughly 28% more open contracts. Rising open interest during a falling price means traders were adding positions into the drop, not merely closing them. That is consistent with leveraged shorts being opened or leveraged longs being added, and the data does not say which. It does show the move was not just holders selling.

The second Broadcom shock in four months

This is not the first time Broadcom has moved Korea. Herald Corp reported on 7 June 2026 that Broadcom's fiscal second-quarter sales (February to April) were $22.19 billion, up 48% from a year earlier but slightly below the $22.27 billion consensus, and that Tan declined to raise the company's AI semiconductor sales guidance for the year. The Kospi closed that week at 8,160.59, down 315.56 points or 3.72%, after touching about 8,930 earlier in the week. Samsung fell 6.40% and SK Hynix 9.92%. The Philadelphia Semiconductor Index fell 10.26% in the US session, and the won weakened past 1,540 per dollar, which Herald said accelerated foreign selling.

Put the two episodes side by side and the difference is instructive. In June the trigger was a number: a guidance decision that told investors something about demand. In October it was a remark with no new data. Yet SK Hynix's one-day fall of about 10% in October matches its 9.92% in the June episode, and Samsung's 7.4% exceeds its 6.40%. Herald's translated article does not make clear whether its stock figures are for the week or the day, and the Kospi figure was a weekly one, so this is not a like-for-like comparison. If anything it understates the point: a single day now matches a period then. A stock that falls as much on an opinion as on a guidance decision is a stock whose price is being set by sentiment about the AI trade.

What the remark does and does not claim

The strongest counter-argument is that Tan was describing investors, not Broadcom. He called expectations extreme; he did not say demand for Broadcom's products had fallen, and nothing in the sources shows a change in Broadcom's orders. The Options Trading Report adds that when asked what could stop the AI buildout, Tan pointed to money, which that outlet read as a warning about credit and AI-linked debt. It also notes the remarks came days after a Wall Street Journal report that Broadcom is arranging more than $50 billion in private financing for a custom AI chip it is building with OpenAI. We have not verified that report. If it is accurate, the context is that a supplier was commenting on financing risk while helping arrange financing, which is not a contradiction but is worth knowing.

Did Broadcom's own AI demand change? There is no evidence in the sources that it did. Tan's statement can be true of the market and irrelevant to Broadcom's order book, and markets have repeatedly treated the two as the same thing.

The CXMT overhang

A second story ran in parallel. The Options Trading Report said China's CXMT reported “key progress” on a new memory-cell architecture to boost density, with a server-memory product planned by the end of 2026, and that the news was less extreme than early trading chatter suggested, with yields, performance and customer qualification unresolved. Separately, The Information reported on 31 August, as relayed by The Standard, that CXMT has begun producing HBM3E in small quantities, with scale-up planned for 2027 and Alibaba's T-Head and Cambricon testing the parts; no yields or volumes were given. We have not re-read The Information's report.

Is CXMT a real threat to memory pricing? The honest answer is that the reported evidence is about intent and early production, not market share. A new competitor in high-bandwidth memory would matter to Samsung and SK Hynix precisely because their valuations assume scarcity. But small-batch production without disclosed yields is far from a pricing event, and the market's reaction was ahead of the evidence. The CoinDesk report linked the slide to both AI-investment worries and Chinese chip competition without separating their contributions, and nobody can.

What would change the read

• Foreign investor flows and the won: in June, Herald tied currency weakness to foreign selling. The October sources give no currency data, so this remains unchecked.
• Broadcom's next earnings report, which would show whether the remark reflects anything in its order book.
• Evidence of CXMT yields or customer qualification for high-bandwidth memory, rather than architecture announcements.
• Whether the open interest build in SK Hynix derivatives unwinds quickly, which would suggest positioning rather than a repricing of fundamentals.

The evidence shows a large single-day fall in two stocks and an index heavily exposed to them, a remark with uncertain date and meaning, and a precedent from June. It does not show that AI demand has changed, that the move reflects anything beyond sentiment and positioning, or what happens next.