Introduction: Hidden Worries Behind Prosperity
August 15, 2026. After a week of volatile consolidation, sentiment in the Hong Kong tech sector has turned subtle again. Although major giants delivered impressive financial reports in the first half, under the independent observation of "AI Intelligence Bureau," a signal that cannot be ignored is emerging: the global boom in AI infrastructure construction seems to be shifting from "short supply" to "structural oversupply." With the release of production capacity by hardware giants like NVIDIA and the implementation of aggressive data center expansion plans by major cloud vendors, Q3 2026 may mark a critical turning point for the AI computing power market. For investors, seeing through the performance divergence of HK cloud service providers to discern the real logic behind "computing power oversupply" will be the key to winning in the next stage of investment.
From "Chip Scramble" to "Inventory Accumulation": The Rapid Reversal of Supply and Demand
Looking back at 2024 to 2025, the main theme of the global tech world was undoubtedly "computing power anxiety." At that time, GPUs were hard to come by, and companies spared no expense to hoard computing resources to seize the opportunity for large model training. However, entering 2026, this situation is undergoing a fundamental change.
According to the latest industry supply chain research data, global shipments of high-performance AI chips increased by 40% quarter-over-quarter in Q2 2026. However, the overall utilization rate of data centers did not see a synchronous significant increase. This "scissors difference" phenomenon is particularly evident in the Hong Kong stock market. As a crucial pole of global cloud computing, HK-listed cloud service providers invested huge capital expenditures (CapEx) in the first half to purchase AI servers and build liquid-cooled server rooms. But as Q3 began, the market started to worry whether these heavy asset investments could be converted into corresponding cash flow returns in the short term.
This concern is not unfounded. Independent analysis suggests that the supply-demand mismatch in the computing power market is mainly reflected in two levels: one is the oversupply of general-purpose computing power, and the other is the partial idleness of high-end training computing power. As the demand for large model training hits a phased peak, and the explosive growth of inference demand has not yet been fully realized, the market is in a "digestion period" of transition. For HK cloud vendors, this means they must face the dual challenge of sharply increased hardware depreciation pressure and the risk of declining rental prices.
HK Cloud Business Performance Divergence: The Moats of Giants and the Survival Dilemma of Small and Medium Enterprises
Under the shadow of "computing power oversupply," the performance of HK cloud service providers shows a significant K-shaped divergence trend. This divergence is not accidental but the inevitable result of corporate strategic positioning and technical barriers.
First Tier: Value Release of Full-Stack Capabilities
Represented by Tencent Cloud, Alibaba Cloud (though mainly listed in the US, its business has a profound impact on the HK stock ecosystem) and the Huawei Cloud-related industrial chain, these head vendors, despite huge investments in hardware procurement, have successfully converted computing costs into high-value-added AI solutions by leveraging their powerful PaaS (Platform as a Service) and SaaS (Software as a Service) ecosystems.
The moat of these giants lies in the closed loop of "Computing Power + Model + Scenario." For example, the industry large model matrix launched by Tencent Cloud in Q2 2026 has penetrated core fields such as finance and healthcare. Customers no longer simply purchase GPU computing power but purchase intelligent decision-making services based on computing power. This model allows head vendors to maintain stable profit margins through software subscriptions and value-added services even if computing power rental prices decline. In the Hong Kong stock market, targets with such full-stack capabilities are often regarded as "safe havens," and their risk resistance has been recognized by capital.
Second Tier: The Dilemma of Pure "Shovel Sellers"
In contrast, some small and medium-sized HK cloud service providers lacking self-developed model capabilities and vertical industry scenario accumulation are facing severe survival tests. Their business models rely heavily on "sublessor" style computing power sub-leasing, that is, purchasing hardware resources at low prices and then reselling them to AI startups at a markup.
However, as major companies cut prices to seize market share and hardware costs are rapidly amortized, the profit space of these intermediaries is being compressed to the limit. The Q3 2026 earnings season is expected to reveal this cruel reality: the revenue growth of cloud service providers lacking core competitiveness will slow significantly, and there may even be an expansion of losses. For such stocks, investors need to remain highly vigilant because "computing power oversupply" is not an opportunity for them, but a direct operational crisis.
Independent Market View: Beware of the "Infrastructure Trap," Seek "Application Layer Dividends"
As an independent market observer, "AI Intelligence Bureau" believes that there is a certain misconception in the current market's perception of AI computing power. Many investors still use traditional manufacturing logic, believing that capacity expansion equals performance growth. But in the AI era, computing power is not the only scarce resource; in some dimensions, high-quality data and efficient algorithms are more important than pure computing power.
We propose the concept of "Infrastructure Trap" to warn investors: do not be confused by the huge capital expenditures of cloud vendors, and must deeply analyze the Return on Investment (ROI) behind these inputs. If the increase in computing power cannot bring a significant improvement in model training efficiency, or cannot be converted into commercially deployable applications, then this part of the investment is inefficient, or even harmful.
Under this logic, the focus of investment should shift from "hard infrastructure" to "soft applications." In the Hong Kong stock market, companies focusing on AI Agents, vertical industry SaaS, and data element processing may hold more investment value than pure computing power providers. Because with the abundance or even excess of computing power, the innovation cost of the application layer will be greatly reduced, which will bring opportunities for explosive growth to companies with unique scenarios and data.
Deep Analysis: The Arrival of the Inference Era and the Rise of Edge Computing
Besides the application layer, the structural changes in computing power demand also deserve key attention. Q3 2026 marks AI's full transition from the "Training Era" to the "Inference Era." With the popularization of generative AI in terminal devices such as mobile phones, PCs, and cars, the demand for edge computing is showing exponential growth.
This puts forward new requirements for the HK semiconductor and consumer electronics industrial chains. Unlike the pursuit of ultimate computing power in cloud training, edge inference values energy efficiency ratio and cost control more. Therefore, HK technology companies with layouts in ASIC (Application-Specific Integrated Circuit) design, computing-in-memory technology, and on-device AI optimization are expected to see valuation restoration in Q3 and the second half of the year. This is not only a technological iteration but also a concrete manifestation of market style shifting from "grand narratives" to "performance realization."
Investment Strategy Suggestions: Finding Certainty in Divergence
Facing the complex market environment of Q3 2026, how should investors layout? "AI Intelligence Bureau" proposes the following three independent suggestions:
- Prefer head cloud merchants with "Computing Power Monetization" capabilities: Focus on leading enterprises that can effectively pass on computing costs and have a continuously increasing proportion of AI-related revenue. Although their short-term valuation is constrained by the macro environment, in the long run, they are the biggest beneficiaries of the AI dividend.
- Avoid mid-stream enterprises with pure computing power leasing: For companies lacking technical barriers and relying solely on reselling computing power, it is recommended to avoid them. Against the backdrop of "computing power oversupply," their performance elasticity is extremely poor, and they are prone to becoming the main force for valuation destruction.
- Lie in wait for dark horses in on-device AI and vertical applications: Actively look for small and mid-cap growth stocks with landing applications in edge computing chips, AIoT (Artificial Intelligence of Things), and specific industries (such as financial risk control, autonomous driving). These companies are less affected by the cloud computing power price war and are expected to achieve a Davis Double Play in performance under the background of the explosion of on-device AI.
Conclusion: Rationally View "Oversupply," Embrace Value Reassessment
In summary, the "computing power oversupply" alarm sounded in Q3 2026 is not the end of the AI industry, but a necessary weeding out of the false to keep the true. It marks that the AI industry is saying goodbye to wild growth and entering the deep waters of refined operation and commercial landing.
For the Hong Kong stock market, this is both a challenge and an opportunity. Market fluctuations will squeeze out bubbles, allowing real value targets to surface. As investors, we should abandon the mentality of blind following and hype, return to the essence of business, and deeply analyze the core competitiveness of enterprises. Today, when "computing power" is no longer scarce, what is scarce are outstanding enterprises that can use computing power to create real value and solve practical problems. This is the core insight that "AI Intelligence Bureau" hopes to convey to every investor through independent commentary.
