CITIC SEC has released a research report stating that recent news of Meta Platforms Inc (NASDAQ: META) considering leasing out part of its computing power has once again sparked market concerns about a potential oversupply. This, combined with concentrated worries over cloud providers' cash flow pressures, the sustainability of upstream price increases, slowing Capex growth, and overly crowded AI trades in recent months, is the primary reason for the significant volatility in technology stocks.
In the short term, the core of Meta's move is to revitalize its existing, older computing power assets. Considering that it continues to develop advanced models and invest in next-generation computing hardware, the plan to lease out capacity does not contradict continued investment in new computing power. Furthermore, as computing power rental rates are still rising recently, concerns about an oversupply are unfounded. This round of adjustment in tech stocks is more a process of deleveraging and rebalancing by funds amid the recent global tightening of liquidity, rather than a reversal of the AI industry trend.
For the medium to long-term narrative, close attention must be paid to whether the coming months will see another "out-of-the-box" AI capability moment similar to the start of the year with OpenClaw and Coding Agent. The market may still be underestimating the potential of models like Mythos/Fable 5. The ability of domestic and international models to handle long-context, complex tasks continues to improve rapidly. In terms of commercialization, AI for Science (AI4S), knowledge work, and physical AI are all expected to form a relay, but validation and waiting for a "jump moment" are required. Regarding Capex, the surge in AI-related bond issuance this year is expected to support capital expenditure in 2027.
Additionally, it has been observed that as overseas AI assets enter a phase of high crowding, high correlation, and high volatility, international capital has begun seeking differentiated sources of returns again. The domestic computing power "Plan B," which possesses differentiated value, remains resilient and is expected to attract foreign capital allocation. With the earnings season approaching, it is recommended to focus on sub-sectors with high certainty of strong earnings growth and reasonable valuations.
In terms of industry trends, the report recommends domestic FABs and semiconductor equipment with a positive narrative, as well as the optical communication sector with lower valuations. On the price increase chain, segments with high AI exposure and early price increases show higher earnings realization, such as memory and upstream PCB materials. The main views are as follows.
Meta's Narrative: Core Focus is Efficiency, Computing Power Shortage Unchanged
Recent media reports about Meta considering leasing part of its computing power have caused market concern. In fact, according to SemiAnalysis, Meta has already contracted over 5GW of capacity in the first half of the year, estimated to correspond to approximately $250-300 billion in expenditure, showing no signs of oversupply. Furthermore, referencing the cooperation models and contract values of SpaceX with Anthropic and Google, Meta's leased computing power, due to stronger cluster capabilities, could generate revenue per MW that is 3-4 times that of peers and can operate on a rolling short-term contract model. Therefore, under this model, Meta can also balance its own model catch-up efforts with returns on its computing power investment.
The ultimate demand for computing power depends on token usage. Meta's entry into the computing power rental market essentially reflects a further concentration of inference demand towards leaders like Anthropic and OpenAI, representing a structural change rather than an overall demand surplus.
Sustained High Capex Growth; Surge in AI Bond Issuance to Support 2027 Capex
In Q1 2026, the four major Cloud Service Providers (CSPs) once again raised their full-year capital expenditure guidance, totaling $700 billion, and provided a positive outlook for 2027. Overall, global tech giants have clear intentions to expand computing power production, and AI infrastructure construction is still accelerating. However, aggressive Capex pressures operating cash flow, introducing high volatility risks.
Against the backdrop of cash flow pressure, the bond market has become the most important supply line. In H1 2026, the five major hyperscalers collectively issued over $130 billion in bonds. According to the Dallas Fed, the issuance of AI-related investment-grade bonds in 2026 is forecast to be around $300 billion. As AI bond issuance explodes, the AI narrative is deeply penetrating from the stock market into the fixed-income market, becoming a structural variable in the US Treasury market. However, tech stocks are highly sensitive to interest rates. Competition for long-term funds between AI bonds and US Treasuries could push up long-term rates and weaken the valuation cushion for tech stocks, thereby suppressing sector valuations. Financing is unlikely to be the main constraint on Capex growth, but the pace of bond issuance's impact on tech valuations warrants attention.
Model Side: Scaling Continues Unabated, Capabilities in Handling Long-Context Complex Tasks Keep Improving
Model capabilities in handling complex, long-context tasks continue to improve. According to METR, Claude Mythos Preview's ability to handle such tasks has already exceeded the 16-hour testing limit. The empirical rule of model long-context task duration doubling every seven months has not slowed. Claude Fable 5 and GPT-5.6 preview show significantly improved coding and agent capabilities compared to the previous generation. Meanwhile, improvements in long-context complex task capabilities are spilling over into cybersecurity abilities, leading to further regulatory tightening. Fable 5 reopened on July 1st after a ban, while the preview version of GPT-5.6 is currently only open to about 20 government-approved trusted partners and organizations.
Price-wise, GPT-5.6 Sol output is priced at $30 per million tokens, slightly lower than Fable 5's $50 per million tokens. Domestic models are further accelerating, with GLM-5.2 becoming the domestic open-source SOTA model and its coding capability already ranking in the global top tier.
Forward Outlook: Earnings Validation Window Opens, Focus on Certainty Themes
With the concentrated disclosure period for US Q2 earnings and A-share interim reports approaching, market pricing power will return to earnings delivery. The current market movement is more driven by profit growth. Continued earnings surprises will strengthen the global growth consensus and push the market to new highs. Short-term pullbacks instead provide layout opportunities for leading companies with earnings support.
For US stocks, the capital expenditures and guidance of the four major CSPs are the core coordinates for observing global AI industry trends. In Q1 2026, their combined full-year Capex plans already reached $700 billion. If Q2 reports maintain high guidance and 2027 growth expectations persist, it will further solidify the logic of the AI computing power industry chain. NVIDIA Corp's (NASDAQ: NVDA) data center revenue, as a direct reflection of Capex implementation, combined with the mass production and delivery of the Vera Rubin platform, constitutes a key window for validating industry prosperity.
For A-shares, core segments of the AI industry chain show outstanding earnings elasticity. The market has shifted from "broad AI concepts" to high-growth, strongly certain sub-sectors. Interim report earnings will become the core pricing factor. It is recommended to focus on leading companies with strong earnings certainty and a good match between valuation and growth rate, while avoiding high-valuation targets based purely on concepts without earnings support.
Computing Power Hardware: Meta Narrative Does Not Alter Upward Trend; Focus on High-Growth, Reasonably Valued Segments
1) Growth Chain: FAB and equipment narratives are positive; optical modules maintain low valuation advantages. Due to the major trend of increasing localization rates and the reality of insufficient domestic computing power construction, the prosperity of domestic computing power is relatively high. Among these, the wafer foundry segment benefits from rising capacity utilization, and the semiconductor equipment segment benefits from memory and advanced node capacity expansion, coupled with localization rate upgrades. Related companies' Q2 2026 earnings have already shown initial growth expectations. With new products like 1.6T optical modules and NPO driving industry trends upward, leading optical module companies still maintain characteristics of high growth and low valuation. Of course, due to high consensus expectations, there may be short-term risks related to order timing fluctuations.
2) Price Increase Chain: Spreading to power and analog segments; segments with high AI exposure and early price increases show higher realization. Reviewing signals from various segments on the price increase chain in June, the most密集 signals came from upstream PCB materials and passive components. Memory and optical fiber/cable maintain high prosperity. Power and analog semiconductors, as well as semiconductor upstream materials, are beginning to see activity. In summary, the magnitude and transmission节奏 of price increases still follow the pattern of AI exposure. The development of AI power supplies and related segments deserves attention, and segments with distribution channels are more prone to price increases. Segments with high AI exposure and early price increases show higher earnings realization and deserve attention during the earnings season. Memory and upstream PCB companies generally showed faster Q2 earnings growth.
Key Risk Factors
Risks include a pullback in high-valuation stocks; slower-than-expected progress in macroeconomic recovery; risks of related industrial policies falling short of expectations; slower-than-expected progress in corporate core technology and product R&D; slower-than-expected AI application adoption; cloud provider capital expenditure falling short of expectations; uncertainties in global and domestic pandemic situations; risks of domestic government and corporate IT spending falling short of expectations due to sluggish macroeconomic growth.
Comments