Global semiconductor stocks slid on Tuesday after Broadcom chief executive Hock Tan likened large swathes of artificial-intelligence investors to gamblers in a casino, deepening market anxiety over debt-fuelled AI spending and intensifying competition from Chinese chipmakers.
The selloff was sharpest in Asia. South Korea’s chip-heavy Kospi index tumbled 7% as Samsung Electronics and SK Hynix — the world’s two largest memory-chip makers — dropped more than 8% each. Japan’s Kioxia Holdings shed over 12%, a gauge of Chinese technology stocks fell 3.6%, and shares of TSMC, Asia’s most valuable company, slipped as much as 2%. US chipmakers had already slumped overnight, extending a rout that has now erased a third of the Kospi’s value since its June record high.
The trigger was an unusually blunt intervention from one of the AI boom’s biggest beneficiaries. Speaking at an event in Singapore on Monday, Tan — whose company’s AI chip business has tripled in two years — told investors that “a lot of investors have no clue what they are investing in,” adding that they set “very extreme expectations only to find out that it’s almost like they are in a casino.”
What Tan said in Singapore
Tan’s remarks, first reported by Bloomberg, were aimed squarely at the money pouring into frontier AI developers — the OpenAIs and Anthropics of the world — where he argued enthusiasm has outrun any real grasp of the underlying economics. He warned that financing the AI buildout, not chip shortages, may become the industry’s key bottleneck: a striking claim from a man whose own company is amassing tens of billions of dollars to fund custom chips for those same developers.
He did not stop at investors. Tan also took aim at how enterprises are spending on AI, comparing corporate budgets for AI “tokens” — the units of text that AI services charge for — to children burning through arcade credits: money spent fast, with little to show for it. His prescription was that AI model developers need to find customers who can generate real returns from deploying the technology, rather than relying on users who are, in his view, wasting money.
The comments carry unusual weight precisely because of their source. Broadcom is not a sceptic warning about a bubble from the sidelines; it is one of the largest direct beneficiaries of the spending Tan now describes as reckless. In September, the company raised its full-year AI revenue guidance to about $58 billion — up 186% on the prior year — and told investors to expect that figure to roughly double to $115 billion in fiscal 2027 and double again to $230 billion in fiscal 2028. Its AI semiconductor revenue hit $16.7 billion in the third quarter of fiscal 2026 alone, more than tripling year on year, on custom-chip deals with Google, OpenAI, Anthropic and Meta.
Asia takes the hit: a 7% Kospi rout
Korean equities bore the brunt of Tuesday’s selloff. The Kospi’s 7% plunge pushed the benchmark below its 200-day moving average, a widely watched technical threshold, and left it down a third from the record high reached in June. Foreign investors have accelerated their exit, selling more than $6.3 billion of Kospi shares on a net basis in October alone. The Seoul exchange briefly halted Kospi program trading for the first time since August — an emergency brake that underscores how disorderly the selling became.
The falls are all the more striking because the underlying business news has been good. Samsung reported record earnings last week, yet the stock failed to get any lift — a sign, analysts say, that the market is no longer willing to price chipmakers on trailing results. “Samsung’s strong earnings have been overshadowed by concerns about future competition from CXMT and the sustainability of AI spending,” said Jung In Yun, chief executive of Fibonacci Asset Management Global.
Ha SeokKeun, chief investment officer at Eugene Asset Management, said “uncertainty over whether the upcycle can last into 2028 and beyond appears to be weighing on valuations and share prices” of Korean chipmakers. As KCM Trade’s Tim Waterer put it, traders are no longer treating good news or lofty valuations as gospel — they are digging into the numbers and the underlying business case, with the whole AI trade under the microscope.
The CXMT wildcard
Tan’s casino comment was not the only spark. A report from the South China Morning Post on Monday said China’s CXMT plans to roll out a new server-memory product by the end of 2026 that enables higher storage density — an architecture Samsung and SK Hynix are also developing. The report resurfaced fears of rapid technological progress by Chinese chipmakers that have rattled global rivals all year.
The fear is not that CXMT matches Samsung or SK Hynix today, but that it compresses the industry’s famously lucrative memory cycle tomorrow. Memory chips are a commodity business at heart: when new capacity floods in, prices collapse, and the super-profits of the upcycle evaporate. Investors who bought Korean chipmakers as a leveraged play on endless AI demand are now confronting the possibility of a supply shock from the one competitor Western export controls were supposed to hold back.
Yun at Fibonacci cautioned that CXMT “still needs to translate technological progress into commercially competitive production at scale” — but in a market already primed for doubt, the headline was enough.
The paradox at the heart of the trade
There is a deep irony in Tan’s intervention that markets have not missed. The man calling AI investors gamblers runs a company whose own fortunes are now lashed to the same casino. Broadcom is reportedly amassing $60 billion to fund chips for Anthropic, and its credit default swaps — the cost of insuring its bonds against default — printed a record 136 basis points last week, more than triple their level in May, as nearly $500 billion of 2026 AI borrowing gets repriced across the market. Oracle’s CDS hit a record 261 basis points alongside it.
In other words, equity investors are still paying up for AI growth while credit investors are demanding ever-higher compensation for the debt funding it. Goldman Sachs’ derivatives desk estimates that the Magnificent Seven now account for about 22% of total US prime brokerage exposure — the most since early 2022 — that semiconductor exposure has doubled to roughly 12% since the start of the year, and that about $115 billion sits in leveraged semiconductor ETFs. JPMorgan’s factor team puts semiconductor sentiment above the 99th percentile. Those are the footprints of a crowded trade.
The tension was captured neatly by former fund manager Michael Burry, who shared the Bloomberg report on Tan’s remarks with his followers — a signal that the market’s most famous bubble-hunter is watching the same fault lines.
Why nerves are raw right now
Tan’s comments landed on already frayed nerves. Last week saw the collapse of a planned offering by Firmus Grid, an Nvidia-backed data-centre company — an early sign that public markets are getting choosier about funding AI infrastructure. Days earlier, an FT report put OpenAI’s revenue run-rate at about $50 billion rather than the widely cited $70 billion, a gap Goldman’s technology desk attributed to gross-versus-net accounting — but one that raised uncomfortable questions about how much of the AI equity complex rests on unaudited numbers from private companies.
The calendar offers little respite. Earnings from ASML and TSMC this week will give investors their next hard data on how long bumper profit growth in the chip equipment and foundry businesses can last. With expectations this elevated, even solid results may not be enough — as Samsung discovered last week.
Not everyone on Tan’s own stage shared his gloom. Intel chief executive Lip-Bu Tan told the same Singapore audience that hardware would stay scarce for two to three years, while Hock Tan himself insisted demand will outstrip supply for at least two more years. The industry’s physical reality — genuine shortages of leading-edge capacity — remains the bulls’ strongest argument. The question the market is now asking is whether the financial engineering built on top of that reality can survive higher rates and harder scrutiny.
What to watch next
Three things will determine whether Tuesday’s rout is a wobble or a turning point. First, the ASML and TSMC earnings calls: any softness in forward guidance on AI-related orders would confirm the market’s fears. Second, the pace of debt-funded AI deals: with credit markets already charging record spreads on the biggest borrowers, each new financing tests how much leverage the boom can carry. Third, CXMT’s progress toward commercial-scale production of its new memory architecture — the variable that could turn a demand story into a supply glut.
For investors caught between the AI boom and its growing fragility, this is a moment to stress-test assumptions rather than chase momentum. Equity volatility of this kind is also a reminder of why diversification into non-correlated assets matters: when chip stocks swing 8% in a day, many investors rebalance toward hard assets — track live gold prices to see how the classic hedge is responding. And with debt-funded deals facing record credit spreads, anyone considering leverage of their own should first run the numbers on our loan and EMI calculator — because if borrowing is getting expensive for Broadcom, it is getting expensive for everyone.
FAQs
What exactly did Broadcom’s CEO say?
Speaking at an event in Singapore on Monday, Hock Tan said “a lot of investors have no clue what they are investing in” and that they set “very extreme expectations only to find out that it’s almost like they are in a casino.” He also warned that financing the AI buildout may become a bigger problem than chip shortages, and compared enterprise spending on AI tokens to children burning through arcade credits. Notably, he remains bullish on physical demand, saying it will outstrip supply for at least two more years.
Why did South Korean stocks fall the most?
The Kospi is one of the world’s most chip-concentrated major indices, with Samsung Electronics and SK Hynix dominating its weighting — so a semiconductor selloff hits Seoul harder than anywhere else. Both stocks fell more than 8% on Tuesday, the index dropped 7% and closed below its 200-day moving average, and the exchange briefly halted program trading for the first time since August. Foreign investors, who have sold $6.3 billion of Korean shares net in October, accelerated the decline.
Is Broadcom itself exposed if AI spending slows?
Deeply — which is what makes Tan’s comments so striking. Broadcom guides to about $58 billion in AI revenue this fiscal year, doubling to $115 billion next year and $230 billion the year after, built on custom-chip deals with Google, OpenAI, Anthropic and Meta. It is reportedly raising $60 billion to fund chips for Anthropic, and its credit default swaps hit a record 136 basis points last week — a sign bond investors are charging heavily for the risk. If AI capex disappoints, few companies have more to lose.
Who is CXMT and why does it matter?
ChangXin Memory Technologies (CXMT) is China’s leading DRAM maker and the great white hope of Beijing’s semiconductor self-sufficiency drive. A South China Morning Post report said it plans a new higher-density server-memory product by end-2026 — technology Samsung and SK Hynix are also developing. Analysts caution CXMT must still prove it can manufacture at commercial scale, but memory is a cyclical commodity business: credible new supply is enough to threaten the pricing power behind the current super-profits.
What should investors watch next?
Earnings from ASML and TSMC this week are the immediate test — forward guidance on AI orders matters more than backward-looking beats, as Samsung’s record results showed. Beyond that, watch credit markets: record CDS spreads on AI borrowers like Broadcom and Oracle show bond investors growing wary of debt-funded capex, and any further widening would raise the cost of the entire AI buildout. In volatile stretches like this, reassessing portfolio balance — including hard-asset hedges you can track via our gold rate page — beats chasing the bounce.
🤖 This article was rewritten by Feed and Figures' editorial AI from a report originally published by Financial Times. Facts and quotes are preserved from the original; the rewrite focuses on clarity and structure. For the unedited original, see the source link below.
