“Agent AI” rises, will CPUs make a comeback?
```
Over 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 workflows, the underlying computing power demands of data centers are undergoing a fundamental "dramatic shift."
According to news from Chasing Wind Trading Desk, HSBC Securities pointed out in its latest deep industry report that a new era of "computing power rebalancing" has begun, and CPUs are 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 mainly relied on large-scale GPUs for parallel computation—what we often call "brute force aesthetics." But the workflow of Agentic AI is completely different; it is a "super orchestrator."
HSBC’s research report points out that Agentic AI not only needs to output content, but also completes tasks through Retrieval-Augmented Generation (RAG) and accessing external APIs (such as email systems, web search, collaborative office software). These tasks involve a large number of logic branches, task syncing, and multithreading processing.
Comparative data shows that, in Agentic AI tasks, the parallel advantage of GPUs is reduced, while the flexibility of CPUs in multitasking makes them shine.
According to the latest research from Cornell University, about 44% of computing power in Agentic AI workflows depends on CPUs, which is 3-4 times higher than the CPU share in traditional AI workflows. CPUs are no longer mere system "auxiliaries," but have become the indispensable "brains" of the computing engine.
Why Are Server Shipment Expectations Continuously Surging? Supply Far Less Than Demand
Such structural changes in computing power demand directly translate into strong support for server shipments.
HSBC Securities has dramatically raised its baseline forecast for global server shipment growth in 2026 to 20% year-on-year (up from 4%), well above the market's general expectation of 10%-15%. Even more astounding, HSBC estimates that "potential demand" in the market is as high as 60%.
Why has the market not fully realized these demands?
The main reason lies in supply chain "bottlenecks." The report emphasizes that the current supply gap for CPUs and DRAM memory is as high as 30%-40%, and core components like SSDs and power chips also face a 10%-30% supply shortage.
This means that, over the next two years, global server shipment growth will be a "toothpaste-style" release constrained by supply, and these persistent supply shortages have led to a large backlog of orders. Not only does this ensure shipment logic for 2026-2027, it also makes HSBC confident that the global server growth cycle may extend to 2028.
Capital Expenditure Boom: "Extra Fuel" Provided by the One Big Beautiful Bill Act
What’s driving this hardware frenzy is not only technical upgrades but also policy dividends.
According to HSBC calculations, with the implementation of the US "One Big Beautiful Bill Act," enterprises purchasing data center assets (including servers) can enjoy 100% accelerated depreciation and immediate R&D expense deduction. This provides cloud service providers (CSPs) with extremely strong financial incentives for hardware investment.
HSBC predicts that the combined capital expenditure of major cloud service providers (including Google, Microsoft, Amazon, Meta, and Oracle) will reach $743 billion in 2026, an 81% year-on-year increase.
This capital injection is structurally shifting: in 2025, CSPs will invest about 35% of their capital expenditure in servers; by 2027, this proportion will soar to 59%. This resource tilt from "land and building" to "computing power equipment" precisely targets benefits for the server supply chain.
Who Benefits? The "Rebalancing" Dividend in the Supply Chain
Market style is shifting from a pure "GPU logic" to a more balanced "computing node value chain."
HSBC points out that, as the value share of CPUs in the rack increases, server ODMs (original design manufacturers) and key component suppliers will become the main beneficiaries in this computing power rebalance.
In summary, this computing power arms race has entered "deep waters." In the past, "whoever owns graphics cards wins," but in the future, "whoever can manage computing power allocation holds the key to efficiency."
The resurgence in CPU demand driven by Agentic AI not only injects long-term certainty into the server market, but also draws a clear hardware upgrade logic line for investors.
~~~~~~~~~~~~~~~~~~~~~~~~
The above exciting content is from Chasing Wind Trading Desk.
For more detailed interpretation, including real-time commentary, frontline research, etc., please join the 【Chasing Wind Trading Desk·Annual Membership】
Risk Warning and DisclaimerThe market has risks; investment needs caution. This article does not constitute personal investment advice, nor does it consider the special investment objectives, financial status, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific situation. If you invest based on this, you are solely responsible. ```