The Zhitong Finance App learned that CITIC Construction Investment released a research report saying that Hengke is waiting for the AI narrative to be re-evaluated, and that US stocks are focusing on rebalancing hard and soft. In 2026, Hang Seng Technology ranked lower in the performance of major global stock indexes. On the one hand, overseas macro-liquidity tightened and suppressed valuation repair. On the other hand, the Hengke Index structure was mismatched with trends in China's AI industry. The weight of procyclical assets such as the Internet and automobiles was too high and dragged down by the domestic macroeconomy, while semiconductors and AI software weighed less than 20%, contributing limited to the upward drive of the index. The problem with US stocks is not AI demand, and the cost of failed verification will be amplified by high valuations and high concentration: second-order capital expenditure is negative, depreciation is concentrated, and antitrust is unresolved. The three simultaneously affect the profit quality of technology leaders; the valuation difference between hardware and software has reached the critical point of transition, and the allocation must be anchored in cash flow and safety margins.
CITIC Construction Investment's main views are as follows:
Hengke is waiting for the AI narrative to be re-evaluated, and US stocks focus on rebalancing software and hardware
In 2026, Hang Seng Technology ranked lower in the performance of major global stock indexes. On the one hand, overseas macro-liquidity tightened and suppressed valuation repair. On the other hand, the Hengke Index structure was mismatched with trends in China's AI industry. The weight of procyclical assets such as the Internet and automobiles was too high and dragged down by the domestic macroeconomy, while semiconductors and AI software weighed less than 20%, contributing limited to the upward drive of the index. Looking ahead to 2027, Hengke's profit side is hedged by two forces: the moderate restoration of nominal growth rate and the improvement in the pattern brought about by anti-internal circulation favors the recovery of main business profits, but AI capital expenditure depreciation concentrates on eroding profits and cash flow; the valuation side is suppressed by overseas interest rate hike cycles, and the Internet AI narrative still needs to be re-evaluated. Therefore, it may be difficult to have a large-scale market at the index level; it is more likely to be a structured market. The opportunities focus on rebalancing under extreme valuations, cooling expectations of interest rate hikes, and AI narrative catalysts. On the racetrack, priority is given to sectors with high shareholder returns, vertical e-commerce, gaming and other sectors with weak macroeconomic relationships, and some AI directions. The problem with US stocks is not AI demand, and the cost of failed verification will be amplified by high valuations and high concentration: second-order capital expenditure is negative, depreciation is concentrated, and antitrust is unresolved. The three simultaneously affect the profit quality of technology leaders; the valuation difference between hardware and software has reached the critical point of transition, and the allocation must be anchored in cash flow and safety margins.
AI & Internet: Commercialization extends from programming to a wider range of tasks, and platform value revaluation depends on the realisation of additional profits. Increased model capabilities and declining inference costs will jointly lower the application threshold for agents. Industry competition will further shift from model testing and user acquisition to complex task delivery and payment conversion. Looking ahead to 2027, we believe there are three main lines: 1) The core issue of internet platforms is shifting from “whether they can keep up with model capabilities” to “whether they can turn investment into profit.” On the enterprise side, office agents expand payment requirements from developers to non-R&D positions. Cloud vendors can simultaneously undertake software subscriptions, model calls, and supporting computing power requirements; on the personal side, agents move from information question-and-answer to advertising and transactions, and existing user portals, merchant resources, and payment systems will reduce commercialization costs. Full-stack collaboration determines supply and cost advantages, and the application ecosystem determines demand and monetization efficiency. Together, the two support return on capital. According to the benchmark calculation, the ROICs of Ali IaaS and MaaS were 13.1% and 17.7%, respectively.
Tencent's small amount is about 16.4%, and AI investment has the foundation to generate positive profits. 2) The migration of AI portals does not equal the loss of commercial value from domestic platforms. Data authorization, transaction payment, and offline execution form ecological barriers. It is difficult for external agents to independently complete a closed loop of transactions. The impact of AI on the platform is more likely to be reflected in improvements in recommendation, delivery, and operational efficiency. Increased profits from the main business and new application revenue are expected to jointly support valuation. 3) Model competition will eventually return to intelligent improvement, and end-side penetration will open up space for the next stage. Rapid iteration at the cutting edge of cost performance makes it difficult for simple price reduction to form a lasting advantage. Post-training and infrastructure engineering capabilities determine the upper limit of undertaking high-value tasks.
Consumer Internet: The overall boom is still weak, marginal improvements in the competitive landscape have become the dominant variable, and the stock selection logic has switched from share expansion to profit margin and certainty. On the e-commerce side, consumption is still recovering weakly supported by policies, but the difference in growth rates between platforms has converged, and the second-order sector of the industry continues to improve, and a first-order inflection point may occur in the next 2-3 years; compared to GMV growth, the upward trend in the monetization rate is a more important variable in 2027. Cross-border e-commerce still has room to expand, but the US and Europe have successively abolished small tax exemptions and superimposed ESG legislation, the cost advantage of direct mail has been systematically narrowed, and the industry has moved to the three paths of compliance, localization, and branding. The policy is the core disturbance. E-commerce is one of our most optimistic directions: pharmaceuticals, MRO, and second-hand 3C are all in the middle and early stages of institutional expansion. The low penetration rate determines that the growth rate is driven by category expansion rather than share seizure. Competitive factors are category depth and compliance assets rather than subsidy intensity, and profit growth is systemically faster than revenue growth. The mitigation of the local lifestyle pattern has mainly benefited from competitive strategic adjustments such as Ali and Byte and increased AI investment. Meituan CLC has reversed losses, and takeout UE has been corrected, and is financially sustainable; however, Keeta is probably the biggest expectation gap in 2027: the market expects it to enter a harvest period and reduce losses in new businesses, and going overseas is Meituan (03690)'s firm strategic direction. Brazil's two-tier operation and potential expansion in Southeast Asia means that investment may or increase.
Media Internet: The key word for media Internet in 2026 is redistribution in deceleration: advertising of short videos is slowing down at the same time as e-commerce, AI videos are moving from model competition to price wars and commercialization, and music and short dramas are experiencing a new round of fragmentation. Facing 2027, we recommend paying attention to three major clues: 1) Whether the main cash flow of the main business can be maintained (fast): The main short video industry decelerated in 2026. The core reason was macroeconomic weakness, e-commerce and advertising markets themselves decelerated, and e-commerce tax and live streaming regulations put pressure on the pressure. At the same time, AI investment also brought additional profit erosion. 2027 focuses on the inflection point of the main business; 2) Multi-modal video models: 2026 is a year where AI video crosses the producible threshold, and the production time and reference material are greatly increased. Success rate increased by more than 9 times , production bottlenecks shift from generation to scripting, IP, and delivery. In terms of model capabilities, the pattern converged to 2+2. Byte Seedance and Kuaishou (01024) Keling dominate performance and revenue; MiniMax (00100) H3 breaks through the price anchor with open source+2K+ low price, and Alibaba Wanxiang 3.0 enters the productivity entrance with omnipotent reference+document input, and the overall ranking of domestic models has already entered the first tier; 3) inflection point in the content platform landscape (soft drinks, red fruits, green bubbles): In terms of music platforms, the marginal improvement of competition is expected in 2027, but the inflection point is still necessary. In terms of skits, 2027 is expected to slow down the growth rate and diversify the structure, moving from AI sales to boutique+, bringing the biggest variables to the entry of Tencent's Green Bubble.
Physical AI & intelligent driving: Recently, the main line of physical AI has gradually switched from mass production of humanoid robots to a closed loop of world models and data, but the real progress is still hardware first, intelligence later remediation. The landmark change in the last two months is the first time that the world model has the three product forms of “interactive, reconstructible, and operable” at the same time. Looking ahead to 2027, the core points are: 1) the world model has entered the product year, but the general world model has not yet settled; 2) the robot brain moves from upper body capture to full-body mobile operation and generalization, but it is still an excellent demonstration in controlled scenarios; 3) industrialization comes first, followed by intelligence. In terms of smart driving, L2 will enter a strong supervision cycle in 2026. At the same time, L3 will move from pilot to limited commercialization. Responsibility boundaries, entry mechanisms, and standard systems will begin to be formed, and L4/Robotaxi will enter the large-scale verification stage under the resonance of China and the US. Based on the inflection point of the smart driving industry, the industrial chain segment circuit robotaxi, smart driving chips, and lidar are all expected to fully benefit.
Major cloud vendors in North America: In 2027, the supply constraints of North American cloud vendors will shift from GPUs to electricity, high-bandwidth memory, and delivery capacity. Investors should allocate assets according to the return of computing power per unit and cash conversion capacity. Amazon (AMZN) relies on AI load to drive CPU cloud demand, combined with high-margin advertising business, and has the strongest cash conversion path. Microsoft (MSFT) maintains steady B2B commercialization performance and operating cash flow through large-scale penetration of Copilot and corporate authority governance barriers. Alphabet (GOOGL) achieved simultaneous expansion of search and cloud businesses, but free cash flow turned negative in a single quarter, and subsequent valuation expansion depended on revenue conversion and free cash flow recovery from the $707 billion procurement commitment. Although advertising efficiency has improved, Meta (META)'s capital expenditure accounts for 51% of revenue and is compounded by litigation expenses. The free cash flow safety cushion is the thinnest, and it is necessary to wait for the free cash flow inflection point to occur.
US stock software sector: Investments in the US stock software sector should shift from conceptual hype to infrastructure platforms with cross-system process data, high net income retention rate (NRR), and high free cash flow profit margin (in line with Rule of 40). ServiceNow has a cross-system action portal, and 50% of the new business has completed non-seat billing migration; SAP's underlying data barriers are deep, benefiting from the 2027 cloud migration cycle; Workday payment conversion is lagging; and Salesforce suggests low configuration due to slowing organic growth and transfer of the interface layer. Snowflake (NRR of 126%, directly benefiting from private data governance and inference consumption) is preferred for data infrastructure, and MongoDB is the target of highly flexible transactions.