CPU Makes Comeback in AI Era as Demand Surges

What’s behind the ‘CPU renaissance’ and where China stands in designing and manufacturing central processing units in the AI age

· CPUs Making a Comeback in AI

Once overshadowed by GPUs, CPUs are re-emerging as critical infrastructure for the next phase of AI, especially agentic AI systems.

· Intel Betting on a ‘CPU Renaissance’

Intel CEO Lip-Bu Tan says CPUs are becoming the “indispensable foundation” of the AI era. Intel’s data center revenue rose 22% to US$5.1 billion in Q1.

· Why CPUs Are Back in Focus

While GPUs remain vital for AI training and inference, advanced AI agents require:

o Orchestration of tasks

o Memory management

o Data movement and coordination
These functions are primarily handled by CPUs.

· Shift Toward Heterogeneous Computing

AI infrastructure increasingly depends on CPU-GPU collaboration rather than GPU dominance alone.

· Nvidia Also Reinforcing CPUs

Nvidia’s new Vera Rubin architecture increases CPU importance, moving closer to a 1 CPU : 2 GPU ratio versus older designs with one CPU supporting many GPUs.

· Big Growth Forecast for CPU Demand

Morgan Stanley estimates US$32.5–60 billion in additional CPU demand by 2030, driven by agentic AI.

· Intel’s Growth Drivers

o Google collaboration for Xeon deployment

o 18A manufacturing node ramp-up

o Expanded US production capacity

· Competition Intensifying

o AMD expanding EPYC processors

o Arm moving into its own data center chips

o Nvidia strengthening in-house CPU strategy

· Supply Crunch Pushing Up Prices

o CPU prices reportedly up 10–15%

o Lead times stretched from 1–2 weeks to 8–12 weeks

o Foundry, packaging, and advanced-node constraints adding pressure

· Why China Still Lags

China trails leaders like Intel and AMD in both CPU design and advanced manufacturing, partly due to restrictions on EUV lithography and other export controls.

· China’s Multi-Track CPU Strategy

China is advancing through:

o x86-based designs (Hygon)

o Arm-based chips (Huawei)

o Indigenous LoongArch ecosystem

o Open-source RISC-V development

· Recent Chinese Progress

o Hygon developing next-gen server CPUs

o Loongson improving competitiveness

o Alibaba’s T-Head launched Xuantie C950, a high-performance RISC-V processor for AI workloads

· Bottom Line

The AI era is shifting from “GPU-only” thinking to balanced CPU-GPU systems, giving CPUs renewed strategic importance while China works to narrow gaps in chip self-reliance.

[ABS News Service/24.04.2026]

The central processing unit (CPU) – the chip technology that drove Intel’s sales and profits for decades but was overshadowed by the graphics processing unit (GPU) in the AI age – is making a comeback.

That’s according to Intel CEO Lip-Bu Tan, speaking on the company’s latest earnings call. “The CPU is [reasserting] itself as the indispensable foundation of the AI era,” he said. “This isn’t just our wishful thinking, it’s what we hear from our customers.”

Intel shares rose about 20 per cent in after-hours trading on Thursday night local time, after it beat first-quarter earnings estimates, extending its rally since last year.

The chip giant, which appeared to lag in the early stages of the artificial intelligence boom, is showing signs of a rebound, mainly driven by surging demand for CPUs due to increasing AI agentic capabilities. Revenue from Intel’s data centre business rose 22 per cent to US$5.1 billion in the first quarter.

CPUs have returned to the spotlight in the era of agentic AI, marking a shift away from the earlier narrative of “more GPUs, more compute”.

The South China Morning Post explains the logic behind this “CPU renaissance” – and where China stands in design and manufacturing.

What is new in the CPU market now?

Intel, long the dominant player in CPUs, is regaining momentum after losing ground during the early, GPU-driven phase of the AI boom.

The company recently announced a multi-year collaboration with Google to deploy its Xeon processors – Intel’s data centre-focused CPUs – while expanding co-development of custom infrastructure chips.

The ramp-up of its 18A manufacturing node, alongside expanded production capacity in the US, has also underpinned its growth.

Nvidia has also renewed emphasis on its in-house CPU, Vera, as part of its Vera Rubin rack. “The CPU is no longer simply supporting the model; it’s driving it,” said Nvidia CEO Jensen Huang during his GTC 2026 keynote in March.

Rivals are also repositioning. Advanced Micro Devices (AMD) is scaling up its EPYC processors for AI-era workloads, while Arm Holdings is moving up the stack from architecture licensing to developing its own data centre chips. Together, they are competing for a larger share of AI infrastructure spending.

Why did CPUs fall out of favour, and why are they back?

GPUs are better suited to AI training and inference because they can run thousands of parallel operations simultaneously, making them far more efficient at handling the matrix calculations that underpin large language models. This sidelined CPUs in the race to develop AI models.

However, as AI evolves into more complex, agent-driven systems, these workloads require more orchestration, memory management and data movement – functions handled primarily by CPUs.

In a heterogeneous computing architecture, CPUs and GPUs are increasingly interdependent, with overall system efficiency determined as much by CPU capability as by accelerator performance.

Nvidia’s earlier Hopper AI server architectures often featured one CPU supporting more than a dozen GPUs. Nvidia’s Rubin platform, however, is moving closer to a one-to-two ratio.

In a recent note, Morgan Stanley estimated that CPU-side orchestration could account for 50 to 90 per cent of total system latency, and projected US$32.5 billion to US$60 billion in incremental CPU demand by 2030, driven by agentic AI.

What does the current supply-demand picture look like?

The prices of CPUs, especially those designed for AI data centres, are rising as surging demand collides with tight supply across the semiconductor supply chain.

Intel and AMD have reportedly informed customers of price increases of about 10 to 15 per cent, while lead times have lengthened significantly – from about one to two weeks previously to roughly eight to 12 weeks, a Nikkei report in March said.

Foundries are struggling to keep up with advanced-node demand, while chip packaging bottlenecks and lower yields add further strain. Foundry capacity is also being diverted to GPUs and memory chips, exacerbating shortages.

Constraints in key materials and equipment, alongside geopolitical disruptions, have pushed up costs across the industry.

Where does China stand in CPU design and manufacturing?

“China still lags leading players such as Intel and AMD, both in design and manufacturing,” said Joe Fei, a semiconductor analyst at LongBridge Securities, pointing to restrictions on access to advanced equipment such as extreme ultraviolet (EUV) lithography machines under US export controls.

On the design side, China is advancing along multiple tracks: improving performance in established ecosystems such as x86 and Arm – via AMD-derived designs at Hygon and in-house Arm cores at Huawei Technologies – while also pursuing greater independence through home-grown architectures like LoongArch and open-source RISC-V.

RISC-V is an open-standard instruction set architecture (ISA) that enables the development of CPUs under a shared software ecosystem. It became available for chip developers to configure and customise designs under the Switzerland-based non-profit RISC-V International in 2015.

Recent developments include Hygon’s next-generation server CPUs, Loongson chips showing competitiveness in select benchmarks, and Alibaba Group Holding unveiling a high-performance RISC-V processor aimed at AI workloads.

T-Head, Alibaba’s chip unit, in March revealed its Xuantie C950, a RISC-V processor that can be scaled by stacking multiple cores for data centre workloads and was claimed by the company as the most powerful of its kind globally.