Macro analyst Geo Chen said he sold all of his AI semiconductor and infrastructure positions in June and moved to cash, arguing that a crowded equity rally now faces pressure from leveraged unwinds in South Korea, a possible buildout of excess computing capacity, and rising long-term U.S. borrowing costs.
Chen’s view centers on a potential reversal in the trade that has driven technology markets: companies supplying memory chips, data-center equipment and computing power for artificial intelligence. He argued that the sector’s gains have relied on continued spending increases and a belief that proprietary AI models can sustain high prices, assumptions now being tested by cheaper open-weight competitors and signs of softer demand for computing services.
South Korea’s market has become an immediate source of stress in that argument. Retail traders there have used 2x and 3x leveraged exchange-traded funds tied to SK Hynix, the KOSPI and memory-chip exposures, Chen said. Citigroup estimated that the resulting unwind had erased $38.7 billion. Figures cited in the report put forced liquidations at 1.2 million accounts, equivalent to roughly one in 30 South Korean adults.
Leveraged products magnify both gains and losses. When the underlying shares fall, funds may need to reduce exposure quickly to maintain their target leverage, adding selling pressure during a decline. That mechanism can make a regional correction relevant to global semiconductor names when the same suppliers, chipmakers and AI infrastructure companies are held across international portfolios.
Crowded momentum trade faces a weaker backdrop
JPMorgan data cited by Chen placed crowding in S&P 500 momentum leaders at 93.3%, close to a previous peak reached before July 2026. The measure suggests that traders remained heavily concentrated in shares that had already outperformed, even as the South Korean liquidation cycle began to expose vulnerabilities in leveraged AI-related positions.
Crowded momentum trades can remain intact for extended periods, but they become harder to manage when funding costs rise or earnings fail to justify increasingly high valuations. Chen’s concern is less about a single disappointing quarter from a chipmaker than about a synchronized reassessment of AI spending, revenue potential and financing conditions.
That pressure was visible, according to the report, in the market response to larger capital-expenditure plans. Google was said to have increased its 2026 AI capital-spending target to $205 billion from $195 billion, followed by a 7% decline in its share price that day. Such a reaction would indicate that markets are becoming more concerned with the returns generated by AI infrastructure rather than treating higher spending as automatically positive.
The report also pointed to Silicon Data’s SDLLMTK token-spending index, which it said peaked in June before trending lower. If token spending — a measure intended to track consumption of large-language-model services — is weakening while infrastructure budgets continue to expand, data-center operators could face a more difficult balance between capacity additions and actual usage.
Open models challenge AI pricing assumptions
Chen also highlighted rapid progress among Chinese AI developers. Moonshot AI’s Kimi K3 ranked first on Frontend Code Arena with 1,679 points, ahead of Claude Fable 5 at 1,631, according to the figures he cited. Alibaba subsequently released Qwen 3.8, an open-weights model reported to have 2.4 trillion parameters.
Open-weight models make their underlying model parameters available for others to run, modify or deploy. Their advance does not automatically undermine every commercial AI service, since enterprises may pay for reliability, integrations, security and specialized tools. Yet strong performance from low-cost alternatives could reduce the pricing power of companies whose valuations assume that access to advanced models will remain scarce and expensive.
Chen’s argument is that AI capability prices could move closer to the underlying cost of compute. That would place greater pressure on private model developers and data-center providers that funded major buildouts through debt or long-duration capital commitments.
The report described a potential oversupply cycle in which data-center projects are delayed or canceled, capacity agreements are renegotiated, and highly leveraged operators struggle to service debt. It also cited Bloomberg charts showing wider credit-risk signals for data-center and hyperscaler-linked debt, including a rise in an AI ecosystem credit-risk measure labeled SPCX.
Public and private technology valuations were also presented as signs of more cautious sentiment. The report said SpaceX traded at $112, 27% below its IPO price, and that OpenAI had pushed its IPO plans into next year. Those examples reflect the report’s broader claim that enthusiasm around AI-linked private-market valuations may be meeting more demanding expectations around revenue, profitability and funding.
Higher treasury yields add financing pressure
Chen linked the technology risks to a sharp move in U.S. rates. The report said the 30-year Treasury yield rose above 5.2%, its highest level in two years, following a Federal Open Market Committee meeting that left rates unchanged. In the scenario described, the Federal Reserve is chaired by Kevin Warsh, and markets interpreted the policy setting as contributing to a “bear steepening,” where long-term yields rise faster than short-term rates.
Higher long-term yields raise the cost of financing capital-intensive projects, particularly data centers that require large upfront spending before generating steady cash flow. They also make future earnings less valuable in present-value terms, a recurring challenge for high-growth technology shares.
The geopolitical backdrop compounds that concern. Chen cited a U.S. Defense Department estimate that the Iran war had cost U.S. taxpayers $37.5 billion so far. His analysis grouped the conflict with COVID-19, the Russia-Ukraine war, Trump-era tariffs and the Strait of Hormuz blockade as inflationary shocks that could keep pressure on energy prices, commodities and government borrowing.
Crypto remains exposed to a tech-led selloff
The report extended Chen’s caution to digital assets, arguing that major cryptocurrencies can trade closely with technology equities during periods of broad deleveraging. It cited a 0.84 correlation between leading digital coins and the Nasdaq during recent technology-sector declines.
Correlation can change rapidly and does not establish that stock-market margin calls will mechanically force selling across crypto markets. Yet the relationship means a technology-led selloff could reduce risk appetite across both asset classes, particularly where traders have borrowed against volatile collateral.
Chen’s cash move reflects a defensive reading of an AI market that has shifted from celebrating capacity expansion to questioning whether demand, margins and financing can keep pace. The next tests are likely to come from model-service usage, data-center contract activity, credit spreads and whether semiconductor earnings continue to support the industry’s enormous capital-spending commitments.
Worried about leveraged AI unwinds and liquidity risk? Learn how to manage volatility with our risk management strategies.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
