China Securities Co., Ltd. Overseas 2027 Investment Strategy: Hang Seng Tech Awaits AI Narrative Re-rating; US Stocks Focus on Software-Hardware Rebalancing

China Securities Co., Ltd. released a research report stating that Hang Seng Tech awaits AI narrative re-rating, while US stocks should focus on software-hardware rebalancing. In 2026, Hang Seng Tech ranked at the bottom among major global equity indices. On one hand, tightening overseas macro liquidity suppressed valuation recovery; on the other hand, the index structure of Hang Seng Tech is mismatched with China's AI industry trends, with overweight cyclical assets such as internet and autos dragged by the domestic macro economy, while semiconductors and AI software account for less than 20% and contribute limited upside to the index. The problem with US stocks is not AI demand, but that the cost of failed validation will be amplified by high valuations and high concentration: the second derivative of capital expenditure turning negative, concentrated depreciation realization, and unresolved antitrust issues all act simultaneously on the earnings quality of tech leaders; the valuation gap between hardware and software has reached a switching threshold, and allocation must be anchored by cash flow and margin of safety. China Securities Co., Ltd.'s main views are as follows: Hang Seng Tech awaits AI narrative re-rating; US stocks should focus on software-hardware rebalancing In 2026, Hang Seng Tech ranked at the bottom among major global equity indices. On one hand, tightening overseas macro liquidity suppressed valuation recovery; on the other hand, the index structure of Hang Seng Tech is mismatched with China's AI industry trends, with overweight cyclical assets such as internet and autos dragged by the domestic macro economy, while semiconductors and AI software account for less than 20% and contribute limited upside to the index. Looking ahead to 2027, Hang Seng Tech earnings are hedged by two forces: moderate nominal growth recovery and improved competitive landscape from anti-involution are favorable for core business profit recovery, but concentrated realization of AI capex depreciation erodes profit and cash flow; on the valuation side, suppressed by the overseas rate-hike cycle, internet AI narrative re-rating is still needed. Therefore, index-level large-scale rallies may be difficult, and the market is more likely to be structural, with opportunities concentrated in rebalancing under extreme valuations, cooling rate-hike expectations, and AI narrative catalysts. On sectors, we prefer high shareholder return sectors, vertical e-commerce and gaming with weak macro correlation, and some AI directions. The problem with US stocks is not AI demand, but that the cost of failed validation will be amplified by high valuations and high concentration: the second derivative of capital expenditure turning negative, concentrated depreciation realization, and unresolved antitrust issues all act simultaneously on the earnings quality of tech leaders; the valuation gap between hardware and software has reached a switching threshold, and allocation must be anchored by cash flow and margin of safety. AI & Internet: Commercialization extends from coding to broader tasks; platform value re-rating depends on realization of incremental profit. Improved model capabilities and falling inference costs together lower the application threshold for Agents, and industry competition will shift from model trials and user acquisition to complex task delivery and paid conversion. Looking ahead to 2027, we see three main themes: 1) The core question for internet platforms shifts from "whether they can keep up with model capabilities" to "whether they can convert investment into profit." On the enterprise side, office Agents will expand paid demand from developers to non-R&D roles, and cloud vendors can simultaneously capture software subscriptions, model calls, and supporting compute demand; on the individual side, Agents will move from information Q&A to advertising and transactions, and existing user entry points, merchant resources, and payment systems will lower commercialization costs. Full-stack collaboration determines supply and cost advantages, while the application ecosystem determines demand and monetization efficiency; together they support capital returns. Under base-case estimates, Alibaba's IaaS and MaaS ROIC are 13.1% and 17.7%, respectively, Tencent Xiaowei is approximately 16.4%, and AI investment has the foundation to generate positive profit. 2) AI entry point migration does not equal loss of commercial value for domestic platforms. Data authorization, transaction payment, and offline fulfillment constitute ecosystem barriers, and external Agents find it difficult to independently complete the transaction loop; AI's impact on platforms is more likely to be reflected in improved recommendation, ad placement, and operating efficiency, with stronger core business profit and new application revenue expected to jointly support valuation. 3) Model competition will ultimately return to intelligence improvement, and on-device penetration opens the next stage of space. Rapid iteration of the cost-performance frontier makes simple price cuts difficult to form lasting advantages, and post-training and infrastructure engineering capabilities determine the upper limit for undertaking high-value tasks. Consumer Internet: Overall prosperity remains weak, marginal improvement in competitive landscape has become the dominant variable, and stock selection logic has shifted from share expansion to profit margin and certainty. In e-commerce, consumption remains in a policy-supported weak recovery, but the growth gap between platforms is narrowing, Douyin and Kuaishou are slowing faster than expected, the industry's second derivative continues to improve, and the first-derivative inflection may appear in the next 2-3 years; compared with GMV growth, rising monetization rate is a more important variable in 2027. Cross-border e-commerce still has room for expansion, but the US and Europe have successively eliminated low-value tax exemptions and added ESG legislation, systematically narrowing the cost advantage of direct mail; the industry is shifting to three pathscompliance, localization, and brandingwith policy as the core disturbance. Vertical e-commerce is one of our most favored directions: pharmaceuticals, MRO, and second-hand 3C are all in the early-to-mid stage of institutional expansion, and low penetration means growth is driven by category expansion rather than share capture; competitive factors are category depth and fulfillment compliance assets rather than subsidy intensity, and profit growth is systematically faster than revenue growth. Easing in the local life landscape mainly benefits from strategic adjustments by competitors such as Alibaba and ByteDance to increase AI investment; Meituan's CLC has turned profitable and food delivery UE has turned positive, providing financial sustainability; but Keeta may be the biggest expectation gap in 2027: the market expects it to enter a harvest period and new businesses to narrow losses, while overseas expansion is Meituan's (03690) firm strategic direction, and dual-front operations in Brazil and potential expansion in Southeast Asia imply contingent increases in investment. Media Internet: The keyword for media internet in 2026 is redistribution amid deceleration: short video advertising and e-commerce are both slowing, AI video is moving from model competition to price wars and commercialization, and music and short dramas are undergoing a new round of landscape differentiation. Looking toward 2027, we suggest focusing on three clues: 1) Whether core business cash flow can be defended (Douyin and Kuaishou): in 2026, the core short video businesses both slowed, mainly due to weak macro conditions, deceleration in the e-commerce and advertising markets themselves, pressure from e-commerce taxes and livestreaming regulation, and additional profit erosion from AI investment; in 2027, focus on the core business inflection; 2) Multimodal video models: capability boundaries continue to rise, and pricing competition begins: 2026 is the year AI video crosses the producible threshold, with generation duration and reference materials greatly improved, model generation success rate up more than 9x, and the production bottleneck shifting from generation to scripts, IP, and traffic buying. In terms of model capabilities, the landscape has converged to 2+2. ByteDance Seedance and Kuaishou's (01024) Kling occupy dual leadership in performance and revenue; MiniMax (00100) H3 breaks the price anchor with open source + 2K + low price, and Alibaba Wanxiang 3.0 enters the productivity entry point with all-around reference + document input; the overall ranking of domestic models has entered the first tier; 3) Landscape inflection points for content platforms (Qishui, Hongguo, Green Bubble): for music platforms, competition is expected to improve marginally in 2027, but an inflection still requires conditions, with the core being slower growth at Qishui Music. For short dramas, 2027 is expected to see slower growth and structural differentiation, moving from AI volume expansion to premium+, with Tencent's Green Bubble entry bringing the biggest variable. Physical AI & Intelligent Driving: The recent mainline of physical AI is gradually switching from humanoid Siasun Robot&Automation mass production to world models and data loops, but real progress remains hardware scaling first and intelligence catching up later. The landmark change in the past two months is that world models for the first time simultaneously possess three product forms: "interactive, reconstructable, and action-conditionable." Looking ahead to 2027, the core points are: 1) World models enter the product year, but general world models have not yet converged; 2) Siasun Robot&Automation brains move from upper-body grasping to full-body mobile manipulation and compositional generalization, but remain excellent demonstrations in controlled scenarios; 3) Industrialization scales first, intelligence follows. In intelligent driving, L2 enters a period of strong regulation in 2026, while L3 moves from pilots to limited commercialization, with responsibility boundaries, access mechanisms, and standards systems beginning to form, and L4/Robotaxi enters the stage of large-scale validation under China-US resonance. Based on the inflection point in the intelligent driving industry, sub-sectors such as robotaxi, intelligent driving chips, and LiDAR are all expected to benefit fully. Major North American cloud vendors: In 2027, supply constraints for North American cloud vendors will shift from GPUs to power, high-bandwidth memory, and delivery capability, and investors should allocate assets based on return per unit of compute and cash conversion capability. Amazon (AMZN) relies on AI workloads to drive CPU cloud demand and, combined with high-margin advertising business, has the strongest cash conversion path. Microsoft (MSFT), with large-scale Copilot penetration and enterprise permission governance barriers, maintains solid B2B commercialization performance and operating cash flow. Alphabet (GOOGL) achieved simultaneous expansion in search and cloud businesses, but quarterly free cash flow turned negative; subsequent valuation expansion depends on revenue conversion from US$707 billion in purchase commitments and recovery in free cash flow. Meta (META), although advertising efficiency improved, has capex at 51% of revenue plus litigation expenses, leaving the thinnest free cash flow safety cushion, and needs to wait for a free cash flow inflection. US software sector: Investment in the US software sector should shift from concept speculation to infrastructure platforms with cross-system process data, high net revenue retention (NRR), and high free cash flow margin (meeting the Rule of 40). ServiceNow controls the entry point for cross-system actions, and 50% of new business has completed the migration to non-seat-based billing; SAP has deep underlying data barriers and benefits from the 2027 cloud migration cycle; Workday's paid conversion lags; Salesforce is recommended underweight due to slowing organic growth and the ceding of the interface layer. For data infrastructure, Snowflake is the top pick (NRR reaches 126%, directly benefiting from private data governance and inference consumption), with MongoDB as a high-beta trading target.