CITIC Securities: A Preview of the Adjustment of US Semiconductor and AI Computing Power

According to the financial news app Zhitong Finance, CITIC Securities released a research report stating that in the next 2-3 years, continuous investment in global AI computing power is expected to continue. In the long term, it will be necessary to address the commercial loop between AI upstream CAPEX investment and downstream application output. Currently, the global semiconductor industry is only experiencing the upswing cycle for 3-4 quarters, still in the first half of this cycle, and the upward movement of stock prices and duration are far from reaching historical average levels. Additionally, there is optimism for investment opportunities in the US stock semiconductor & hardware sector in the next 12 months, and it is recommended to downplay the short-term market volatility impact and focus on the two core themes of high prosperity and bottoming out recovery.

Recent adjustments in the US stock semiconductor & hardware sector: rate cut expectations, elections, AI concerns, etc.

According to Bloomberg data, as of July 31st, the Philadelphia Semiconductor Index (SOX) has fallen by 17.2% from its recent high, with a cumulative decline of 13% in July, significantly underperforming the market benchmark. On July 17th and July 24th, there were single-day declines of 6.8% and 5.4% respectively, marking the largest single-day declines since October 2022. CITIC Securities believes that the recent adjustment in the US stock semiconductor & hardware sector mainly reflects: 1) trading rate cut expectations, with some funds in the market shifting to rate-sensitive sectors (industrials, small and medium-sized enterprises, etc.), and correspondingly reducing the proportion of positions in sectors that have seen significant gains (technology giants, semiconductors, etc.); 2) trading the US elections, with the market rapidly increasing bets on Trump's return to the White House, and evaluating his policy proposals on tariffs & trade, taxes, internal regulation, etc., and their potential impact on the equity market; 3) trading concerns about the sustainability of AI, with the disconnect between short-term AI upstream investment and downstream output leading to continuous market questioning of the rationality of this wave of AI technology, and quarterly reports from tech giants like Google becoming market catalysts.

Underlying logic of this wave of AI: Scaling Law -> Improvement in algorithm intelligence -> Continuous unlocking of applications.

The complex system and rapid evolution of AI itself make it difficult for CITIC Securities to conduct long-term analysis. Returning to the underlying logic of this wave of AI, CITIC Securities believes: 1) the core constraint of insufficient downstream applications lies in the inadequate capabilities of AI algorithms themselves, and improving the intelligence level and usability of AI algorithms is the root of the solution; 2) based on the theoretical research results of OpenAI and Google DeepMind, the intelligence of large language models comes from "compression is intelligence", and the improvement in model intelligence level mainly follows the Scaling Law, which involves expanding the model parameters, increasing the training dataset, and achieving full training process, which is the continuous stacking of AI computing power and the influx of a large amount of high-quality data. Obviously, before the Scaling Law fails, the leading industry participants are likely to continue along this path, continuously investing in computing power to drive the improvement of algorithm intelligence level, thereby unlocking downstream application scenarios, and ultimately achieving a commercial loop. In the face of this revolutionary industry wave, for tech giants, the risk of insufficient short-term investment will be far greater than the risk of excessive investment AI Computing Power Support: Scaling Law, Balance Sheets of Tech Giants.

CITIC Securities concluded that the effectiveness of the Scaling Law and the balance sheets of tech giants are the core factors determining the sustainability of computing power investment: 1) In terms of the effectiveness of the Scaling Law, in the short term, referring to the structure of the human brain, there is still at least one to two orders of magnitude room for improvement in the parameter scale of large models and training datasets. In the next 2-3 years, the market does not need to worry too much about this issue; 2) Regarding the balance sheets of tech giants, CITIC Securities used Google as an analysis sample (refer to "Forward-looking Research on the US Tech Sector - Quantitative Calculation of the Impact of AI Capital Expenditure on Tech Giants' Financial Reports", July 31, 2024), and based on several simplifying assumptions, CITIC Securities estimated that AI investment would drag Google's profit by around $1 billion in 2024, with the potential to turn into a positive contribution from 2025H2. Therefore, even with a simple linear extrapolation, at least in 2024 and 2025, there is no need to worry too much about the willingness of tech giants to invest and their ability to sustain investment.

US Semiconductor Cycle: Still in the first half of the industry's upturn cycle.

Currently, all sub-sectors of the US semiconductor industry are in an upturn phase, and the simulated inventory levels of the last sector to adjust are peaking in 2024H1. Referring to historical patterns, the US semiconductor cycle lasts for 4-5 years in a single cycle (with a downturn of 4-6 quarters and an upturn of 10-12 quarters). If we start counting from 2023Q4, the current US semiconductor upturn cycle has only gone through 3-4 quarters, still in the first half of this cycle. In terms of market fluctuations, in the past 10 cycles, the SOX index has averaged a 164% increase. In the current cycle, the cumulative increase of the SOX index up to the present is 112%. Looking at the potential for price increases and duration, it is highly likely that the SOX index in this cycle has not yet peaked. Even in an upturn cycle, the stock price trend of the US semiconductor sector is not always upward. Continuous fluctuations, even significant pullbacks, are normal for the sector. Considering the Fed's "data-dependent" monetary policy path, the US election, ongoing concerns and doubts about AI, etc., significant fluctuations in the SOX index are expected to be a inherent feature of the market this year, and we need to respond calmly.

Risk Factors: Risks of significant market fluctuations in election years; risks of slow progress in AI algorithm development and application landing; risks of escalating geopolitical conflicts; risks of unexpected economic downturns in Europe and the US; risks of persistent stickiness in US inflation; risks of core technical personnel turnover in tech companies, etc.

Investment Strategy: Based on a comprehensive analysis of the underlying logic of the current AI technology wave, the balance sheets of tech giants, and other factors, CITIC Securities believes that in the next 2-3 years, continuous investment in global AI computing power is sustainable. In the long term, there is a need to solve the business loop between upstream investment and downstream output. Meanwhile, the current upturn cycle of the global semiconductor industry has only gone through 3-4 quarters, still in the first half of this cycle, with the price increase and duration of the sector far from reaching historical cycle averages. CITIC Securities continues to be optimistic about investment opportunities in the US semiconductor and hardware sectors in the next 12 months, and recommends downplaying the short-term impact of market fluctuations, while continuing to focus on the two core themes of high prosperity and bottoming recovery