I'm LongbridgeAI, I can summarize articles.HSBC Research found that 44% of tasks in "agent AI" workflows rely on CPUs, which is 3-4 times that of traditional AI computing, resulting in CPUs and memory becoming the biggest bottlenecks for data center computing power. The bank significantly raised its global server shipment forecast for 2026 to a year-on-year growth of 20%, and believes that due to supply chain capacity constraints, actual market demand could be as high as 60%, with growth dividends continuing until 2028
In the past three years, the narrative logic of the AI server market has been highly singular: whoever owns more GPUs holds the key to general artificial intelligence (AGI).
However, as AI applications evolve from simple chatbots to "Agentic AI" capable of automatically executing complex processes, the underlying computing power demands of data centers are undergoing a fundamental "transformation."
According to news from the Chasing Wind Trading Desk, HSBC Securities pointed out in its latest in-depth industry report that an era of "computing power rebalancing" has begun, with CPUs once again becoming the decision-making hub of AI data centers.
From "Chatting" to "Working": The Computing Power Earthquake Triggered by Agentic AI
Early AI models primarily relied on large-scale GPUs for parallel computing, commonly referred to as "brute force aesthetics." However, the workflow of Agentic AI is entirely different; it acts as a "super orchestrator."
HSBC's research report indicates that Agentic AI not only needs to output content but also requires retrieval-augmented generation (RAG) and access to external APIs (such as email systems, web searches, and collaborative office software) to complete tasks. These tasks involve a large number of logical branches, task synchronization, and multithreading.
Comparative data shows that in Agentic AI tasks, the parallel advantages of GPUs are diminished, while the flexibility of CPUs in multitasking shines.
According to the latest research from Cornell University, approximately 44% of the computing power in Agentic AI workflows relies on CPUs, which is 3-4 times higher than the CPU share in traditional AI workflows. CPUs are no longer just "assistants" in the system but have become an indispensable "brain" in the computing power engine.
Why Are Server Shipment Expectations Continuously Soaring? Supply Far Below Demand
This structural change in computing power demand directly translates into strong support for server shipment volumes.
HSBC Securities has significantly raised its year-on-year growth forecast for global server shipments in 2026 to 20% (up from 4%), a prediction that far exceeds the market's general expectation of 10%-15%. Even more astonishingly, HSBC believes that the market's "potential demand" could be as high as 60%.
Why hasn't the market fully realized this demand?
The main reason lies in the "bottleneck" of the supply chain. The report emphasizes that the current supply gap for CPUs and DRAM memory is as high as 30%-40%, and core components such as SSDs and power chips also face supply shortages of 10%-30%.
This means that the growth of global server shipments over the next two years will be a "slow release" after being constrained, and this ongoing supply shortage has led to a large backlog of orders, which not only secures the shipment logic for 2026-2027 but also makes HSBC confident that the global server growth cycle is expected to extend to 2028.
Capital Expenditure Explosion: The "Extra Fuel" from the One Big Beautiful Bill Act
Supporting this wave of hardware frenzy is not only technological upgrades but also policy dividends.
According to HSBC's calculations, with the implementation of the One Big Beautiful Bill Act in the United States, companies purchasing data center assets (including servers) can enjoy 100% bonus depreciation and immediate R&D expense deductions This provides a strong financial positive incentive for hardware investment by cloud service providers (CSPs).
HSBC predicts that the total capital expenditure of major cloud service providers (including Google, Microsoft, Amazon, Meta, and Oracle) will reach $743 billion in 2026, an increase of 81% year-on-year.
The direction of this capital injection has undergone a structural shift: in 2025, about 35% of CSP's capital expenditure will be directed towards servers; by 2027, this proportion will jump to 59%. This shift from "land acquisition and building" to "purchasing computing power equipment" precisely locks in benefits for the server supply chain.
Who is benefiting? The "rebalancing" dividend of the supply chain
Market styles are shifting from a purely "GPU logic" to a more balanced "computing node value chain."
HSBC points out that as the value proportion of CPUs in racks increases, server ODMs (Original Design Manufacturers) and key component suppliers will become the core beneficiaries of this computing power rebalancing.
In summary, this arms race for computing power has entered the "deep water zone." In the past, "whoever owns the graphics card wins," but in the future, "whoever can manage computing power allocation well will master efficiency."
The recovery of AI-driven CPU demand not only injects long-term certainty into the server market but also outlines a clear hardware upgrade logic for investors