The hum of servers in a hyperscale data center, usually dominated by the whisper of high-bandwidth memory and the insistent whir of specialized AI accelerators, is changing.
Intel threw a conversational grenade earlier this year, suggesting that the nascent wave of agentic AI—AI that can operate autonomously and perform complex tasks—would necessitate a renewed focus on traditional CPUs. Now, the data is starting to bear this out, not in the flashy headlines about NVIDIA’s latest behemoth, but in the less glamorous, yet utterly critical, upstream semiconductor supply chain. Japanese printed circuit board (PCB) manufacturer Ibiden’s recent forecasts offer a granular look at this shift, projecting a substantial increase in demand for general-purpose server products, a direct proxy for CPU demand, in 2026.
It’s easy to get lost in the narrative of GPUs ruling the AI kingdom. They’re the obvious workhorses for training massive models. But agentic AI, the kind that can plan, reason, and execute multi-step processes without constant human prompting, introduces a different set of computational requirements. Think less raw brute force for matrix multiplication and more sophisticated instruction execution, decision-making logic, and data orchestration—tasks where CPUs, particularly high-performance server-grade variants, still hold significant sway.
Ibiden, a key player in the complex web of semiconductor manufacturing, supplies integrated circuit (IC) package substrates. These aren’t the chips themselves, but the foundational layers that connect silicon to the broader PCB, facilitating communication and power delivery. Their product portfolio spans substrates for GPUs, CPUs, and the increasingly common application-specific integrated circuits (ASICs) designed for AI. The company’s financial disclosures for the fiscal year ending March 31st paint a picture of an evolving AI market.
Why Agentic AI Is Backing the CPU Revival
Ibiden’s revised forecast for its Electronics segment — the one that matters for substrates — now anticipates 330 billion Japanese Yen in revenue for fiscal year 2026 (ending March 2027). This is a tangible upward revision from the previous 310 billion Yen projection, and the growth engine isn’t just the usual AI suspects. The company specifically calls out general-purpose server products and switching IC products as drivers. Within these categories, the general-purpose server products are explicitly linked to the CPU market for server applications.
Why the pivot? Ibiden itself offers a clue: “transitions from training to intelligence,” leading to an expected increase in demand for CPUs for general-purpose servers. This isn’t just about incremental growth; Ibiden is preparing for a significant capacity expansion. The firm projects its production load to reach 1.8 times the 2024 level by calendar year 2026, climbing further to 2.4 times by 2028. This aggressive ramp-up is fueled by expected growth in ASICs and AI servers, but critically, it includes server CPUs. Meanwhile, demand for PCs, a traditional CPU stronghold, is anticipated to dip—a clear indicator of where the new growth is originating.
This data point is significant because it validates a narrative that has been brewing on the fringes of AI discourse. While GPUs remain indispensable for the heavy lifting of model training, the deployment and fine-tuning of more autonomous AI agents — the kind that will manage complex workflows, act as sophisticated personal assistants, or power advanced robotics — leans heavily on CPU prowess. These agents need to process information, make decisions, and interact with various systems rapidly and efficiently. CPUs, with their sophisticated architectural features for handling diverse instruction sets and managing complex operating systems, are ideally suited for these tasks.
The Substrate Signal: A Forward-Looking Indicator
It’s easy to dismiss a single supplier’s forecast, but Ibiden’s position in the supply chain makes its pronouncements highly instructive. Substrate orders are placed well in advance, and significant shifts in their order books are often bellwethers for broader industry trends. The fact that they’re gearing up for a near doubling of production capacity in just a few years, specifically citing general-purpose server CPUs, suggests that major chipmakers and server manufacturers are making substantial bets on this CPU resurgence. We’re not talking about a slight uptick; this is a projected demand surge.
This trend has broader implications for the semiconductor industry. It suggests that the AI hardware landscape might be more diverse and nuanced than the current discourse often allows. While the latest generation of GPUs will undoubtedly continue to capture headlines, the market for high-performance CPUs, particularly those optimized for complex logical operations and task management, could see a renaissance. This could mean renewed investment and innovation in CPU architectures specifically tailored for agentic AI workloads, potentially challenging the long-held assumptions about specialized AI accelerators dominating every segment of AI computing.
There’s a historical parallel here. Remember the early days of cloud computing? Servers were packed with powerful CPUs to manage a multitude of virtual machines and services. While specialized hardware eventually emerged for certain tasks, the foundational need for strong general-purpose processing power never truly disappeared. Agentic AI, in a way, is creating a new frontier where that general-purpose compute, but at a much more sophisticated and interconnected level, is essential. It’s not about replacing GPUs, but about creating a more balanced and synergistic ecosystem where CPUs play a starring role in the operationalization of advanced AI.
