According to a research report from CITIC SEC, the global tech sector experienced a significant correction over the past month. While the fundamental outlook for the AI industry remains unchanged, a confluence of negative narratives鈥攊ncluding price hike concerns, doubts about capital expenditure returns on invested capital, the disruption of model landscape and token price expectations by domestic open-source models, and interest rate volatility鈥攅rupted simultaneously. This was compounded by overcrowded AI hardware trades and a reversal of leveraged positions, ultimately triggering a market sell-off. Looking ahead to August, the report suggests that for the broader tech sector and AI narrative, easing in positioning and trading congestion, as well as marginal shifts in profit distribution and bargaining power across the value chain, will be key drivers of sector trends in the coming month. The sector is likely to enter a phase of waiting for new narratives, with a rebalancing between tech hardware and software. From a trading perspective, the report advises focusing on areas with favorable positioning, low valuations, and those that have been oversold in this correction. The improvement in open-source model capabilities is driving "intelligence parity" and "token price parity," suggesting attention should first be given to leading internet companies and cloud providers with low valuations and rich AI application ecosystems. After the narrowing of hardware options, the expansion chain is preferred over the price hike chain, with continued focus on domestic equipment and FABs with upward momentum, as well as low-valuation optical communication tracks and select application companies with positive interim results.
Market: Extreme congestion in the tech sector has eased marginally but has not fully cleared. The report believes that overcrowded trading and positioning, combined with the concentrated outbreak of negative narratives, were the core reasons for the major correction in July. From a fund positioning perspective, Q2 public fund heavy positions further concentrated in the tech sector, with hardware as the main allocation direction. Q2 positioning reached a new high of 59.2%, and the overall overweight ratio increased to 19.55%, driven by electronics and communications. In terms of trading volume, TMT sector volume fell from extreme levels in July but remained high, with active trading. Internally, structural differentiation exists, with trading dominated by electronics. On the valuation front, overseas tech index valuations have digested somewhat after adjustments. Hong Kong stock valuations are in a reasonable range, with ample room for internet recovery. Valuations for A-share core tech assets remain at high levels. Regarding leverage, A-share margin trading cooled marginally in July, but TMT margin balance remains at a relatively high level of around 34%. Korean stocks showed a deleveraging trend in July, but overall leverage remains high. Overall, all three indicators have retreated from extremes but are still far from full clearance. The report expects volatility and structural divergence to be the baseline scenario for August, with sector performance more dependent on fundamental realization and capital reallocation pace in specific sub-sectors.
Models & Technology: Slowing second derivative of overseas model ARR exacerbates market concerns, while domestic model capabilities improve and application deployment accelerates. The underlying logic of overseas disruption to the tech narrative is a suspected slowdown in the second derivative of ARR. In July, concerns over a slowdown in Anthropic's ARR growth triggered a market decline, but third-party ARR data contains significant noise, and growth rates had already declined as early as June. The report tends to view the ARR growth slowdown as a short-term catalyst for market sentiment. As the ARR base grows, a slowdown in growth rate was expected, and the key observation point is whether the absolute monthly sequential increase in ARR is decelerating. Simultaneously, domestic model capabilities and application commercialization are accelerating. The release of Kimi K3 achieved open-source SoTA and topped front-end coding benchmarks, raising concerns about deteriorating competitive dynamics for overseas closed-source models and token price declines. On July 16, Kimi K3 went online, achieving the first global top ranking in the front-end coding benchmark Code Arena, marking the best historical performance for a domestic model. The capability gap between Chinese and US models has narrowed to 3-4 months. Technical iteration under computational constraints validates the innovation capability of domestic model companies, which are accelerating breakthroughs in computational constraints through algorithmic innovation. On July 31, DeepSeek V4 Flash went online, with significantly improved Agent capabilities and prominent cost-effectiveness advantages. With only 284 billion total parameters and 13 billion activated parameters, it achieved a capability leap through post-training. In terms of pricing, standard prices per million tokens for cache hit input, cache miss input, and output are $0.0028, $0.14, and $0.28 respectively, continuing the low-cost strategy.
Internet & Agents: Two major rebalancing processes in structure and flows are underway, with earnings bottoming out and valuation upgrading resonating. Structurally, with the recovery of North American hyperscaler model commercialization and ROIC expectations, global capital is shifting from the input side of AI infrastructure to the output side of cloud and applications. In terms of flows, foreign capital is underweight China and internet stocks, while domestic capital allocation is also at low levels. The sector faces a dual low in fund flows, providing clear upward space as rebalancing progresses. Fundamentals and valuations are expected to rise together. On the earnings front, pessimistic expectations have been relatively well priced in. According to Visible Alpha consensus estimates, the combined Non-GAAP net profit of major internet companies for Q2 2026 is expected to decline 8% year-on-year, a significant narrowing from the 32% decline in Q1 2026. Net profit growth rates for Q3 and Q4 2026 are expected to further recover to +12% and +46%, respectively, with core business cash flow supporting sustained AI investment. On the valuation front, the gap between domestic and global flagship models is continuously narrowing, the Chinese AI narrative is becoming clearer, and overseas funds are likely to revalue Chinese AI assets. Opportunities exist across three segments: Cloud is the first stop for AI demand realization, with intensified competition in the model layer and shortened lead times, as value migrates from single models to cloud and middleware distribution capabilities, and the path of capital expenditure conversion to revenue and profits becomes clearer. Models are the core of intelligent supply, serving as an anchor of confidence for revaluation. Applications are the vehicle for commercialization, with leading players intensively integrating Agent middleware layers. The report estimates that the replacement of professional IDEs by natural language interfaces could expand the Agent market from the narrow coding market of approximately $200 billion to broader knowledge work, with AI suppliers' addressable market reaching approximately $1.5 trillion.
Price Hike & Expansion Chain: New capacity to be released in 2027, price peak not yet achieved, subsequent stock prices expected to see fluctuations. Reviewing historical storage prices, SK Hynix's profitability and stock price, during the two previous price hike cycles from 2016 to 2019 and 2020 to 2023, the company's net profit peak occurred approximately two quarters after the price peak, while the stock price peak occurred within approximately one quarter after the price peak. Therefore, determining the price peak also means determining the stock price peak. The price peak is achieved through demand slowdown and new capacity release. Without considering the growth in AI demand, capacity expansion across various links of the price hike chain is currently concentrated starting in Q2 2026. Considering the typical 1 to 1.5 years for capacity expansion, capacity release is expected to be concentrated in Q2 to Q4 2027.
Performance & Narrative: FAB and equipment narratives are upward, while optical modules maintain low valuation advantages. Due to the major trend of increasing domestic substitution rates and the reality of insufficient domestic computing infrastructure buildout, domestic computing power is experiencing relatively high景气度. Within this, the wafer foundry segment benefits from rising utilization rates, while the semiconductor equipment segment benefits from storage and advanced process expansion, coupled with domestic substitution rate upgrades. With the upward momentum in景气度 driven by new products such as 1.6T optical modules and NPO, leading optical module companies maintain characteristics of high景气度和 low valuations.
Risk Factors: Risks include a pullback in high-valuation stocks, industrial policies not meeting expectations, slower-than-expected progress in core technology and product research and development, slower-than-expected AI application deployment, and lower-than-expected capital expenditure from cloud providers.
Comments