<对话 Tom Lee：这轮韩股暴跌是强制去杠杆，不要在结构性趋势中做波段>

Compiled & Edited: Deep Tide TechFlow

Guest: Tom Lee, Co-founder and Head of Research at Fundstrat Global Advisors, CIO of Fundstrat Capital (managing GRNY ETF, ~$5B AUM), Chairman of the Board of BitMine Immersion Technologies (BMNR)

Host: (Global Money Talk, recorded live at the NYSE)

Podcast Source: Global Money Talk

Original Title: Tom Lee: "We are Close to the Bottom"

Air Date: July 27, 2026

Conflict of Interest Disclosure: BitMine Immersion Technologies (BMNR), where Tom Lee serves as Chairman, holds approximately 5.78 million ETH, making it the largest institutional holder of Ethereum globally. Lee also serves as CIO of Fundstrat Capital, managing the GRNY ETF (approx. $5B), which holds positions in stocks discussed in this episode, including Robinhood. Lee's personal wealth is highly correlated with the price of ETH and the performance of GRNY. All opinions expressed in this episode regarding Ethereum and the crypto market are aligned with his significant financial interests. Furthermore, Fundstrat's core business model is paid research subscriptions; Lee's public statements serve a client acquisition/marketing function. Listeners are advised to consider these relationships in their assessment.

Tom Lee is one of Wall Street's most steadfast bulls. His firm, Fundstrat, sells research to hedge funds and family offices in 26 countries monthly. The GRNY ETF he manages has consistently outperformed the S&P 500 since its launch. He is also Chairman of BitMine, the world's largest corporate ETH holder. This episode, recorded live at the NYSE, comes at a time when South Korea's Kospi has plummeted in a single month, AI semiconductor stocks are experiencing a flash crash, and the market is debating whether the "AI bubble has burst." Lee's core argument is brutally simple: the current sell-off is driven by forced deleveraging of leveraged capital, not a fundamental turning point. He invokes Cisco's history of being cut in half four times between 1993 and 2000 before eventually rallying 100x, arguing we are not even in the late stages of the current AI cycle.

Key Takeaways

Tom Lee believes the past month's crash in the Korean stock market and AI semiconductor sector is a "forced deleveraging" event. South Korea's market has seen a significant introduction of leveraged products in recent years, amplifying volatility in both directions. Data from prime brokers at major US investment banks shows that hedge funds dumped tech long positions at the fastest pace in nearly a decade, suggesting "weak hands" have been washed out. He quotes Peter Lynch and Charlie Munger to support his core advice: don't trade in and out of a secular trend. "Money is made by sitting, not by trading."

Lee uses Cisco's history from 1993 to 2000 to illustrate AI's current phase. Cisco experienced multiple drawdowns exceeding 40% during that cycle, each time accompanied by declarations that "tech stocks are finished." Yet it ultimately rose from $0.80 to $80, a 100x gain. The key difference, he argues, is that the top in 2000 was a true bubble where buyers were fiber-optic companies using unrealistic DCF models to justify purchases. "Today's buyers are the hyperscalers, serious companies buying equipment, racking it, and placing massive orders." Regarding the impact of Chinese AI models like Kimi K3, Lee acknowledges this is an "existential question above my pay grade." However, he points out that open-source models are essentially generic drugs, and someone still needs to pay the R&D costs. He is more focused on the opportunity in AI downstream sectors (Mag 7, software, crypto), which are starting to outperform the upstream semiconductor space. His GRNY ETF has outperformed 92% of its peers this year by sticking to this framework. For the macro outlook in the second half of the year, Lee bets inflation will come in lower than expected (the oil price shock has peaked, housing and wages are weakening), which will force the Federal Reserve to pivot dovish.

Highlights and Key Quotes

Forced Deleveraging, Not the End of the Story

"Hedge funds are selling tech longs at the fastest pace in nearly a decade. The weak hands have been shaken out."

"Every time you get a straight-line move up, it traps the levered longs, and then they get forced to liquidate. That's exactly what's happening now."

"No one can call the exact bottom. But if you sell now and wait for a signal, you'll end up buying back higher."

Sell the Flowers, Water the Weeds: Don't Trade a Bull Market

"Peter Lynch said, if you sell your winners, you're cutting the flowers and watering the weeds."

"Charlie Munger said it best: The money isn't made in the buying and selling; the money is made by sitting."

"If it's a secular theme, you should buy it and forget it."

NVIDIA at 16x PE Isn't Expensive, But Memory Stocks Are Naturally More Cyclical

"NVIDIA's forward PE is 16x, not in the 20s. It has the CUDA moat and a fairly certain upgrade path. It should be re-rated as a 25-30x growth stock."

"Memory and semi-equipment are two steps away from the end customer, carrying the risk of the bullwhip effect: hyperscalers might double-order fearing price hikes, and memory manufacturers could over-expand capacity."

"Cyclical stocks have their lowest PEs at the top of the cycle, which isn't a sell signal. You just need to bet that earnings estimates will continue to rise."

Cisco Cut in Half Four Times, Then Rallied 100x: AI Isn't in the Late Innings Yet

"Cisco rallied 100x from 1993 to 2000. It got cut in half at least four times, and each time people said tech was dead."

"If this were the late stages of an AI bubble, people would be shouting 'this is the bottom, buy the semis!'. What are they doing? They're selling frantically."

"In 2000, Cisco traded at 200x PE, and buyers were fiber-optic companies doing 10-year DCFs at a 6% discount rate. Today's buyers are hyperscalers. They aren't hippies digging trenches to lay fiber."

Ethereum vs. Bitcoin: Yield-Bearing Asset vs. Digital Gold

"Bitcoin is a store of value. The ecosystem wants it to ossify into digital gold. Ethereum is a yield-bearing asset, with a staking yield around 3%."

"BitMine's current staking income is about $6 million per week, $300 million annually. If ETH goes to $5,000, it's close to $1 billion per year."

"Our perpetual preferred shares only require $30 million in dividends annually. Staking income alone covers that."

Catalysts for Crypto are Lining Up: CLARITY Act + Robinhood Chain + Institutional Adoption

"Since the end of June, Ethereum has outperformed memory stocks by 72 percentage points. Someone who lost 40% in memory could be up nearly 30% in Ethereum."

"Robinhood Chain was built on Ethereum, not another chain. Daily volume has already surpassed $1 billion. Robinhood could make $1 billion a year from this chain alone."

"The CLARITY Act is on the one-yard line. It would create a single federal regulator for the entire crypto economy. Japan and Russia have already passed similar legislation. The US must catch up."

Gold Isn't Out of Favor, It Just Rallied Too Much and Needs a Breather

"Gold's rally over the past 3 years is a 5-sigma event in the context of its 12-century history. It certainly needs to digest those gains."

"In an AI world, gold's role as a store of value and safe haven won't change. I recommend everyone hold a little, maybe 1%."

South Korea's Crash: Leveraged Products Amplified Volatility, But The Story Remains Intact

Host: On June 22nd, South Korea's Kospi peaked near 9,300 points, coinciding with the peak of the Philadelphia Semiconductor Index. The past month has been a genuine crash. People are wondering: why this initial parabolic surge followed by a sudden reversal? Are we following the Kospi, or is there a global fear that the AI trade is overheated?

South Korea has performed exceptionally well over the past few years. The underlying logic is that the semiconductor and memory content per unit of global GDP is constantly rising. This means South Korea, both as an economy and a stock market, will be far more important over the next decade than it was in the previous 30 or even 50 years. Earnings should be strong.

However, over the past few quarters, the Korean market introduced a significant amount of leveraged products, which amplified two-way volatility. When the market moved straight up, it trapped leveraged longs, and now they are being forced to liquidate. This is what's happening now: a forced deleveraging. But it doesn't mean the underlying story is over. I believe this correction will prove to be one of the best buying opportunities for semiconductor and AI stocks. By extension, the Korean stock market, AI stocks, memory, and semiconductors will eventually reach much higher highs than before.

Host: It's hard to believe before seeing it, but what signal are you waiting for to confirm the correction is over? Or is this the opportunity, and one should be buying in tranches?

Firstly, timing the market was never a good strategy. If you hold these stocks, you should continue to hold. If you sold and are now waiting for a signal to get back in, you will most likely end up chasing the market higher. No one can call the exact bottom.

However, the signal you'd want to see – a massive deleveraging – has already happened. If you look at prime brokerage data from US investment banks, hedge funds dumped their tech long positions at the fastest pace in three years, probably the fastest in ten years. They have already undergone a significant deleveraging event. We've also seen some very high-profile forced liquidations in South Korea. So, those who were forced sellers ("weak hands") are likely out. My assessment is that we are quite close to a bottom.

But people will still make mistakes trying to pick tops and bottoms. The ones who make the most money are those who just hold. Peter Lynch famously said: Selling your winners is cutting the flowers and watering the weeds. Charlie Munger said it best: "The money isn't made in the buying and selling; the money is made by sitting." If this is a secular theme involving more semiconductors and memory, you should buy it and forget it.

NVIDIA's 16x PE vs. Memory Stocks' 4.5x PE: Not a Direct Comparison

Host: I remember you saying in a previous conversation, "Bears sound smart, but bulls make money." I want to ask about valuations. What is NVIDIA's forward PE? I estimate it's around the 20s?

Actually, it's 16 times.

Host: 16 times, indeed... Their earnings expectations are very strong. On the other hand, SK Hynix was around 7x at its high and has now fallen to 4.5x. Can we directly compare the two?

NVIDIA has proven it has a degree of recurring revenue due to the CUDA platform and a fairly certain upgrade path that compels people to keep buying. It should be re-rated as a growth stock; I think a fair valuation is between 25 and 30 times.

As for memory and semiconductor equipment, they are two steps away from the end customer, creating the risk of the bullwhip effect. Simply put, these sectors are more cyclical because they lack pure order visibility. Imagine the end market is a consumer using an AI lab service, like ChatGPT or DeepSeek. They subscribe through an AI lab, which uses a hyperscaler. The hyperscaler buys chips from NVIDIA, and NVIDIA orders from its suppliers. The suppliers are too far from the end user. In this chain of transmission, a lot of double ordering can occur. If hyperscalers anticipate memory and chips will get more expensive, they might place double orders now to lock in prices. Then, memory manufacturers might over-expand capacity for the future.

This has happened in every cycle, and there's a risk it happens again. So, it's normal for more cyclical names to have lower PEs at the top of a cycle; you should expect multiple compression. But that's not a sell signal. You just need to bet that earnings estimates will continue to be revised upwards. In a machine-to-machine world, robots will need far more memory and storage: Humans eat and have nervous systems, robots need memory and storage. The economy is becoming more memory-intensive and semiconductor-intensive. So I believe estimates will continue to go up. But don't compare the PE of memory stocks to NVIDIA's.

The Cisco History Lesson: Cut in Half Four Times, Ended Up 100x

Host: You said you didn't want to give a long history lesson, but I think your memory of history is very useful for investors. I remember when Cisco had problems, it was partly due to the bullwhip effect: orders collapsed in 2001 because people had over-ordered, and it all fell like dominoes. Is that lesson applicable here? Or is it too early?

The AI story will eventually become a bubble, that's inevitable. Anytime you have a secular demand story and the market underestimates volatility, people make poorly risk-adjusted decisions – underestimating the risk.

But I don't think we are in the late stages of an AI bubble. The reason is simple: As soon as the market started falling, most people declared a top. If this were a true bubble, people would be saying "this is the bottom" and frantically pouring money into semiconductors. But they aren't; they are frantically selling.

Let's look at Cisco. From 1993 to 2000, that's 7 years, but it was essentially one cycle: the internet buildout cycle. Cisco started at $0.80. By 1997, it was at $9, a 10x gain. Then in 1997, it corrected 40% due to the Asian Financial Crisis, and everyone said "Cisco's story is over." What happened? By 1998, it went from $5 to $18, double the previous high. Then 1998 hit with Greenspan's "irrational exuberance" speech, the Russian default, and the LTCM collapse. Cisco fell from $18 to $9, a 42% drop. Many people again declared tech stocks were finished. I vividly remember that time, people dancing on the grave of tech stocks, saying the trade was over. Then Cisco rallied from $9 all the way to $80 in 2000. From 1993 to 2000, it was a 100x gain. From its 1998 high, it took only 18 months to rally 5x.

That's when the real top for Cisco happened. I