The AI computing theme saw a massive rebound in US semiconductor stocks before Thursday's market open, marking a dramatic shift from the prolonged deleveraging and liquidation of crowded positions that had plagued the sector. This surge appears to be a long-overdue, sentiment-driven explosion of bullish energy, breaking the recent pessimistic narrative around AI capital expenditure failing to deliver timely returns.
For strategists who remain cautious on global equities, the past two months have seen investors struggle to find a definitive signal to break the US market's new pattern of violent, directionless oscillation. At Thursday's open, the semiconductor sector executed a powerful reversal from "forced liquidation of leveraged positions" to "aggressive short covering." Lam Research Corp, a super-major in semiconductor equipment focused on 3D NAND high-aspect-ratio (HAR) etching, surged over 20% after reporting better-than-expected earnings and raising its guidance.
When examining the recent results and outlooks from AI supply chain leaders like Samsung Electronics, United Microelectronics Corporation (UMC), TSMC, SK Hynix, and Seagate Technology, a clearer picture emerges. The physical demand linked to AI computing infrastructure hasn't deteriorated in lockstep with the collapse of stock prices and leveraged positions. However, one day's rally in US semiconductor stocks isn't sufficient to confirm a new bull leg for the global AI computing trade. Chris Caso, a strategist at top Wall Street firm Wolfe Research, noted that the Philadelphia Semiconductor Index (SOXX) roughly doubled in three months before falling about 25% from its peak. He views the recent weakness more as a recalibration of expectations after a steep rally. Caso predicts demand for AI chips will outpace supply until at least 2028 and continues to rate NVIDIA Corp as the best AI chip investment, arguing that the feared slowdown in hyperscale cloud capital expenditure hasn't materialized. Instead, the competitive race around AI agents is forcing cloud giants into a "no-choice-but-to-spend" environment.
Goldman Sachs' trading desk highlights the market's lack of "juice," suggesting a gap between a technical bounce and a trend reversal that must be filled by capital flow repairs. According to their metrics, global hedge fund gross leverage sits at the 93rd percentile over the past five years. With the S&P 500 breaking below a key CTA trigger level, continued downside could unleash approximately $15.7 billion in systematic selling over the next week and $68 billion over the next month.
Goldman's trading team also reports cooling retail activity, mutual funds and overseas investors turning cautious ahead of the US midterm elections, and August's seasonally weak capital flows. While corporate buybacks provide a reliable floor, they may not be enough to immediately propel high-beta stocks to new highs.
Stuck in the "Wreckage Rut": CTA Selling Looms, Retail Enthusiasm Fails to Return
According to Goldman Sachs, the market's lack of clear direction will persist. A mix of conflicting factors continues to trap the S&P 500 in its trading range since early June, unchanged even after Wednesday's decline. The primary factor is positioning: although hedge fund and retail leverage is no longer extreme, it indicates little appetite for further aggressive risk-taking. Goldman's senior traders, including Gail Hafif, state that gross leverage among global hedge funds hovers near the 93rd percentile of five-year data. Mutual funds and overseas investors may remain on the sidelines until after the US elections in November. Meanwhile, the positive impact of rising corporate buybacks could be offset by seasonally weak August capital outflows.
In a recent research note to clients, Goldman traders wrote, "There are very limited sources of 'juice' for the market in the near term. We still need to navigate through the wreckage of the past few weeks before we can talk about any meaningful re-risking." This view provides ammunition for skeptics who argue the S&P 500, stuck in a roughly 350-point trading range for nearly two months, is unlikely to break out soon. The benchmark index fell 1.5% on Wednesday to its lowest close since June 10. The Nasdaq 100 fell 2.1%, extending its decline from a record high in June to 11%. The sell-off occurred amid growing concern over the spending pace of US tech giants and the potential for high oil prices to reignite inflation.
Goldman traders indicate the near-term environment is unlikely to ease. Their data shows August is historically one of the weakest months for overall stock fund flows, with net outflows from mutual funds and ETFs matching May's levels as the largest of the year. They expect mutual funds to remain cautious until after the November elections, historically keeping cash on the sidelines before deploying it afterward. Overseas investors also tend to reduce US equity exposure in the months leading up to the vote. Retail investors have also pulled back significantly. Goldman reports that average daily trading activity this month is over 3% lower than the five-year average, indicating another source of equity demand is fading. Data from Vanda Research shows that retail investors sold stocks on Tuesday at the highest level since the pandemic.
Systematic investors, or CTA strategies, represent another source of potential volatility. Goldman estimates that these "fast money" investors, who follow market trends rather than fundamentals, currently hold about $196 billion in US stocks, a moderate level relative to history. However, the S&P 500 is now below a key CTA trigger point. If the market falls further, it could trigger approximately $15.7 billion in US stock selling over the next week and about $68 billion over the next month.
A New Bull Run Brewing in the Wreckage of Deleveraging? Semiconductors Lead the Charge for the Primary Uptrend
For bullish investors seeking relief, corporate buybacks are expected to offer some support. Goldman's buyback desk notes that roughly 31% of S&P 500 companies are currently in their open buyback windows. They estimate this will rise to over 50% by next week and reach 90% by mid-August. Goldman traders wrote in their report, "This will bring one of the market's largest buyers back into the stock market. For the coming August, this is the strongest and most reliable capital flow supporting US stocks."
The semiconductor reversal on July 30 was broad-based, encompassing equipment and storage giants like Lam Research, Sandisk, Western Digital, Seagate, and Micron, alongside AI computing chain leaders like Arm Holdings, Advanced Micro Devices (AMD), Intel, TSMC, ASML, Broadcom, and NVIDIA. This suggests capital isn't just covering individual oversold positions but is aggressively buying the entire AI computing supply chain. The synchronized strength across equipment, memory, CPU/IP, foundry, and interconnect chains doesn't contradict Goldman's view of a market lacking overall "fuel." The constraints of mutual fund caution, cooling retail, high leverage, and potential CTA selling limit the S&P 500's ability to break out broadly. Semiconductors, however, have their own industry-specific catalysts: Lam's strong order guidance, sustained AI capex from Microsoft and Meta, tight memory supply-demand, and extreme short-covering. This looks more like a structural power grab by AI hardware within a broader index range-bound market. The rally has the hallmarks of a powerful reversal, but only sustained breadth and a confirmed technical breakout would signal a true upgrade from "chasing a semiconductor bounce" to a new allocation phase.
This counter-offensive has more robust earnings catalysts than a typical technical bounce. Lam Research reported record June-quarter revenue of $6.72 billion, up 15.1% sequentially, with non-GAAP operating margins expanding to 38.4%. It forecasts next quarter's revenue midpoint to jump to $8.1 billion with adjusted EPS of $2.15. Management explicitly stated AI demand is reshaping wafer fab equipment needs through higher-layer NAND, advanced DRAM, HBM, and complex packaging. Arm Holdings, the provider of the instruction set architecture for consumer electronics and AI data center servers, saw revenue rise 22% year-over-year to $1.29 billion, with data center royalty revenue more than doubling. It expects its self-designed data center CPU to generate over $2 billion in demand over the next two fiscal years. Microsoft Azure revenue grew 43%, with a 45% forecast for next quarter, while maintaining its actual AI infrastructure expansion plans, proving that cloud computing investment is converting into cloud revenue, enterprise orders, and cash flow, not just idle GPU stockpiling.
Wall Street bulls are also seeing their core thesis validated by earnings. Wolfe Research's Chris Caso expects AI chip demand to outstrip supply until at least 2028 and continues to see NVIDIA as the best AI chip investment. He believes the feared slowdown in hyperscale cloud capex hasn't occurred, and the competitive dynamic around AI agents leaves hyperscalers with "no choice but to invest." Wolfe also raised earnings forecasts for NVIDIA and Broadcom, suggesting that future opportunities like NVIDIA's upcoming Rubin system and unbooked business are not fully reflected in current estimates. Caso also believes memory tightness could extend the DRAM and NAND price up-cycle into 2028 or 2029. Bank of America projects global cloud and AI infrastructure capex will approach $1.5 trillion by 2027, growing 40-50% year-over-year. These views converge on a single underlying logic: expanding model size, rising inference concurrency, continuous agent operation, and the evolution from single servers to clusters will simultaneously increase demand for GPUs, custom ASICs, HBM, server DRAM, NAND, advanced packaging, etching/deposition equipment, and high-speed interconnects.
A recent report from Citigroup suggests the AI arms race is shifting from "whose model is smartest" to "who can produce intelligence at the lowest cost and highest efficiency under physical constraints." Open-weight models like Kimi K3 are rapidly approaching the frontier of closed-source models, meaning model capabilities are commoditizing faster. However, the simultaneous explosion in parameter count, long context windows, and multi-step Agent reasoning is shifting the bottleneck from pure FLOPs to HBM capacity and bandwidth, GPU high-speed interconnects, cluster scheduling, and power access. NVIDIA's own research indicates HBM is often the primary scaling constraint as model size, sequence length, and batch size grow. IEA forecasts show AI data center electricity consumption growing significantly faster than overall power demand, while power grid construction cycles are generally longer than data center deployment cycles.
Analysts at Morgan Stanley, led by Brian Nowak, have significantly raised their 2027/2028 capital expenditure forecasts for the world's five largest hyperscale cloud operators (Meta, Amazon, Microsoft, Google, and SpaceX), to approximately $1.2 trillion and $1.4 trillion, respectively. The firm has also revised its 2026 US big tech capex forecast sharply upward to $805 billion from $433 billion a year earlier. Based on Samsung Electronics' closing price of approximately 207,000 won and SK Hynix's closing price of 1,322,000 won on July 30, Nomura's target prices for these AI supply chain leaders imply massive upside. A 670,000 won target for Samsung suggests 225% potential growth over 12 months, while a 4,700,000 won target for SK Hynix implies roughly 255.5% potential growth. Nomura's analysts argue these companies should no longer be viewed as traditional memory stocks dependent on the PC and smartphone cycle but redefined as structural growth assets in AI infrastructure. Their thesis rests on three pillars: AI training, inference, and data center expansion driving demand persistently above supply growth; a resonant super-cycle for HBM and general memory; and their roughly 6x forward P/E ratios, significantly below TSMC's ~20x, failing to fully reflect earnings sustainability and ROE improvement. Consequently, Nomura raised its target price for Samsung from 590,000 won to 670,000 won and for SK Hynix from 4,000,000 won to 4,700,000 won.
Comments