洪灏：AI泡沫交易进入新阶段

Original source: Hong Hao's Macro Strategy

Before the end of July, the market once again caught traders off guard. The Nasdaq recorded its strongest trading day since June, and the Philadelphia Semiconductor Index surged 8% in a single day, nearly recouping the nearly 10% decline of the previous three sessions. Microsoft rose more than 15% on the back of over 40% growth in its cloud business and its first-ever breakthrough past $100 billion in annualized revenue — its best single-day performance since 2008. Microsoft's market value soared by $45 billion in a single day, the largest one-day market cap gain in human history. Meanwhile, Amazon rose another nine points after earnings and raised its full-year capital expenditure guidance to $120 billion. LAM Research surged 18%, its strongest day since 1999. Samsung also delivered record quarterly operating profit while warning that memory supply tightness could persist into 2028.

The success of Microsoft's cloud business in its earnings proved the viability of AI commercialization, validating that massive AI capital expenditure can generate returns. Amazon, meanwhile, raised its capex guidance, yet the market rallied in response — a stark reversal from the previous norm where companies were punished for spending more. Overnight, a market that just a week earlier was debating whether "AI capital expenditure is out of control" suddenly shifted to discussing "when will capacity be sufficient." Although this is based on just two companies so far, since these are two of the most important tech giants, the market's interpretation of their upgraded capex plans carries significant instructive weight.

Yesterday at 3 PM on July 30, I discussed the relationship between the AI technology revolution and the AI stock bubble at a VIP closed-door meeting hosted by Binance. The essence of a bubble is how market prices reflect future earnings expectations. When market price movements deviate far from fundamental support and become extreme, we define this phenomenon as a bubble. Even if the AI bubble bursts, it cannot change the upheaval that the AI technology revolution will bring. And these changes are already happening all around us.

During the Q&A session, an audience member asked me about her ETF position — a 2x leveraged ETF on a Korean memory company — which was deeply underwater, and asked what she should do. I replied that I only make directional predictions but do not advise on trades, and that this ETF should "surge 50% or even higher around tomorrow." I also shared this conclusion with readers on my Planet before the market opened this morning. As of the time of writing, this ETF has surged a stunning 70%. The Korean index soared 18% today, yet remains down for the week. This sudden, unexpected market move also demonstrates the importance of how the detailed data charts, analysis, and logical reasoning in my exclusive reports complement the short, fast-paced commentary and technical charts on my Planet.

Chart 1: My market call made at the closed-door meeting at 3 PM on July 30.

Here, I am not seeking credit, but rather hoping to share the difficulty of trading in an era of epic volatility.

In times of epic turmoil, correctly predicting direction is fundamentally different from surviving with your positions intact. Even if you predict the direction correctly, you may not make money, because what determines trading success is never the destination, but the path. Many who shorted the Nasdaq in 2000 died in the final short squeeze of 1999 — they were ultimately right, but didn't survive long enough to be proven so. The decay of leveraged products, margin calls, and redemptions triggered by drawdowns are all costs of enduring volatility in such turbulent markets. Often, the market refuses to return to rationality until traders have been liquidated. Many believe that missing out is more painful than losing money. Yet we've all seen the heartbreaking video of Korean retail investors losing everything, but we've never heard of anyone jumping off a building because they missed a rally.

Therefore, position management matters more than the prediction itself at this moment. Because your position determines whether you can survive the game and reach the finish line. The truly appropriate approach is to scale in gradually or leave some cash on the sidelines, rather than betting heavily all at once in response to massive market swings. Better to make less than to be knocked out. What matters is not pinpoint accuracy, but asymmetric odds: when wrong, you only lose time value; when right, you capture the entire trend.

Ultimately, trading at this moment is about shifting from "prediction" to "response." As Sun Tzu said, those who were skilled in battle in ancient times first made themselves invincible, then waited for the enemy to become vincible. Being invincible depends on oneself; victory depends on the enemy. In the twilight of a bubble, one must first ensure they cannot be defeated.

So, is this the turning point in the AI bubble trade?

If we look at the leading stocks and compare their gains against their fundamentals, then if the AI bubble trade has entered the phase of separating truth from falsehood, we should see high-quality companies leading the rally. At this point, capital shifts from indiscriminate buying to differentiating between true leaders and story-driven companies — the market moves from a phase where anything touching the AI narrative rallies to a phase of separating wheat from chaff.

However, our data analysis shows that in this rebound, companies with weaker fundamentals (using gross margin as a proxy indicator) rebounded more strongly. The magnitude of rebounds is proportional to the preceding declines. Therefore, we cannot yet conclusively say that this round of AI bubble trading has entered the phase of separating truth from falsehood. This looks more like a beta rebound, a short squeeze, rather than a quality rally. On the same day Microsoft surged, Meta fell nearly eight points due to disappointing earnings and soft guidance. From this data point, the market does appear to be starting to differentiate between "can capex translate into visible revenue" narratives. But we believe one or two data points cannot tell the full story.

Meanwhile, traders should also note that in last night's technical rebound, US stocks, US Treasury yields, and gold all rose simultaneously. The 10-year yield rose rather than fell, breaking above 4.66% — the highest since 2007. Stocks, bonds, and gold rising together has never been the picture of normalized risk appetite; that is a market still pricing term premium and fiscal dominance, forcing a smile amid AI earnings strength. A market that truly believes "the problem is solved" would not have shown only technology and consumer discretionary sectors rising in last night's massive rally while other sectors fell. This narrow market participation precisely indicates tightening liquidity conditions at the margin, and it is precisely for this reason that we see the "when the whale dies, everything flourishes" pattern on the tape.

There was another important market event last night — the trigger for the rebound — yet it was almost completely overlooked by the market.

Japan's Ministry of Finance intervened in the currency market during the New York session, pushing the yen from its 40-year low of 164 down to the 157 range, a single-day decline of nearly three percent — the largest one-day move since the end of 2022. The market broadly expected intervention to come after the Fed and Bank of Japan policy meetings, but the MoF deliberately struck between the two meetings to catch everyone by surprise.

However, this intervention has not yet triggered a blowup in yen carry trades, because the US simultaneously intervened in the market as well, ensuring that the yield spread between US and Japanese bonds did not narrow due to the FX intervention. This preserved the profit margins of yen carry trades and helped stabilize the market. As a result, the Nikkei surged today alongside other Asian markets despite the sharp yen appreciation. Historically, unilateral yen intervention has a poor track record, but coordinated US-Japan intervention is a completely different story: the 1985 Plaza Accord and the 1998 joint US-Japan yen buying both changed the yen's trend at those times. Of course, the scale and persistence of those coordinated interventions bear no comparison to last night's one-night stand. In April and May this year, the Bank of Japan's intervention burned through roughly $70 billion to no avail, because the US-Japan rate differential widened rather than narrowed, causing further yen depreciation. If Japan wants to intervene further, US coordination and Treasury backing would be needed for Japan to smoothly sell US Treasuries and buy yen. Such operations would further push up long-term rates and further tighten market liquidity. If the US Treasury yield curve flattens further alongside tightening liquidity conditions, US stocks are likely to continue reflecting tightening liquidity at the margin and slowing US economic growth.

Chart 2: US stock performance reflects yield curve flattening

The largest US bank believes Korea's deleveraging is nearing its end. Liquidation of Korean leveraged ETFs is largely complete, hedge fund leverage on Korean equities has been reduced by approximately 90%, and foreign selling has also slowed markedly — with 90% of the $110 billion net outflow over the past year coming from just two memory heavyweight stocks. The KOSPI has fallen more than 40% from its June 22 high, with a forward P/E of just 5x and price-to-free-cash-flow also at 5x, leading the bank to conclude that Korea's deleveraging process should be largely finished.

However, we believe the decline in Korean leveraged ETF assets under management can be almost entirely explained by NAV depreciation, while net redemptions from these funds are near zero, meaning fund units have not changed. Meanwhile, cumulative inflows into leveraged products remain near elevated levels and have not turned to outflows. The report's language is "inflows stalled," and stalled inflows are a different matter from outflows. If fund units remain intact, it means leverage has not exited the market — it has merely been repriced. And 2x leveraged products will mechanically re-lever on the way up. Any sharp rebound will rebuild the same reflexive structure in its original form. As such, it is difficult to conclude that the deleveraging process of leveraged ETFs is complete. We can only say that leverage in the Korean market has been significantly marked down, and the forced-liquidation phase has likely passed. A reassessment of leverage levels can occur if the Korean market declines again without triggering forced liquidations and a downward death spiral.

So, can this rebound in chip stocks continue?

We currently view this rebound as a tradable rebound. The first wave of indiscriminate selling has likely largely come to an end. The market's interpretation of Microsoft's and Amazon's earnings and capex plans suggests the market is beginning to believe that large-scale capex programs can be commercialized. However, the circumstances of these two companies may not be universally applicable. Last night's US stock rebound saw participation only from technology and consumer discretionary, while other sectors fell. And within the technology sector, the data shows that lower-quality companies rebounded more sharply. As such, this rebound looks more like a beta rally than a quality-differentiated rally. Only when the market begins to differentiate on quality and AI stocks no longer all rise together will we see a sustainable market trend.

Chart 3: Korea's 40-day collapse of nearly half — forced liquidation phase has likely largely stabilized.

The bursting of the internet bubble in 2000 also featured a rotation. Capital did indeed retreat from internet companies with clearly fake earnings and rotated into names everyone agreed were "solid" at the time — Cisco, Intel, Nortel, Sun Microsystems. At the time, these companies all had real revenue, real earnings, real supply shortage narratives, and an impeccable "selling shovels" logic. Yet ultimately, these stocks all lost 80% to 90% of their value. The flight to quality within a bubble is a phase of the bubble: it gives the feeling of managing risk, but in reality it concentrates risk exposure into the part of the market with the highest capital expenditure leverage. Today's memory sector, relative to other market segments, is in a position similar to Nortel back then.

If this rebound can further show several signs — such as improvement in market breadth rather than just a bounce in deeply oversold low-quality names; a pullback in long-term US Treasury yields rather than stocks, bonds, and gold all rising together; and most importantly — market declines no longer triggering chain reactions — only then can we say that the market has truly reversed.