CPUs Take Center Stage in AI Earnings

As inference and agentic AI deployments rise, CPUs could equal and even eclipse GPU demand. That's the finding emerging from recent chipmaker announcements and earnings calls.

CPUs are required to support the sequential logic typical of AI agents, and their capabilities are essential to preprocess data for use in AI models. Add to that the capability to orchestrate and ensure optimal deployment of GPUs, and the role of CPUs becomes even more essential.

In AMD’s recent Q1 earnings call, CEO Lisa Su clearly outlined the situation and expressed increased expectations for future CPU revenue:

“In server, we delivered our 4th consecutive quarter of record server CPU revenue. Revenue increased more than 50% year-over-year…. Inferencing and agentic AI are increasing the need for server CPU compute, as these workloads require additional CPU processing for orchestration, data movement, and parallel execution, in addition to serving as the head nodes for GPUs and accelerators. As a result, we are seeing both stronger near-term demand and deeper engagement with customers on long-term capacity planning.”

CEO Su noted that CPU compute requirements for agentic AI have raised AMD’s estimates of CPU server TAM growth to “greater than 35% annually, reaching over $120 billion by 2030.” This was up from an 18% annual growth estimate AMD issued in November.

AMD plans to release a new Venice series of EPYC CPUs later this year that includes Verano, a CPU built specifically for high-performance computing (HPC) and AI infrastructure, including agentic AI. Meta is set to include Venice and Verano chips in its planned purchase of 6 GWs of AMD chips announced earlier this year.

Investors seemed pleased with AMD’s CPU announcements. In a note this week, analyst Blayne Curtis of Jefferies wrote:

“The [AMD] Server CPU TAM raise was notable, reflecting the structural uplift from agentic AI that is viewed as largely incremental to the GPU TAM…. Server CPU momentum continues with guidance for >70% Y/Y revenue growth in 2Q, and we believe AMD is well-positioned to take further share through Venice (2H26) and beyond.”

Amazon, Arm, and Google Concur on CPUs

Speaking of Meta, the hyperscaler is also a partner and co-developer with Arm Holdings on a new Arm AGI CPU, which Arm characterizes as its first production silicon product designed for agentic AI. Arm says this CPU will perform 2X better than current x86-based environments and allow hyperscaler and enterprise capex reductions in the billions of dollars.

In a letter to shareholders this week, Arm declared that it has seen over $2 billion in customer demand for the Arm AGI CPU for fiscal 2027 and 2028 (which correspond to the calendar years for Arm). And the company acknowledged the increased importance of CPUs in AI workloads:

“Agentic AI is reshaping data center architecture. As AI inference workloads move from human-based queries to continuous, agent-driven workloads, CPUs are critical in managing the orchestration, data movement, memory, security, and workload coordination required by agentic AI. Data centers are expected to require more than 4x current CPU capacity per gigawatt as agentic AI scales, creating a market opportunity of more than $100 billion by 2030.”

On the downside, Arm won’t be delivering its AGI CPUs until 2027. In a note to investors, financial analyst Sebastien Naji of William Blair wrote this week:

“While investors have looked past near-term fundamentals toward a compelling CPU opportunity on the horizon, meaningful revenue contribution will not begin until the end of fiscal 2027, limiting near-term upside.”

Amazon too has highlighted the role of CPUs—and Meta as customer—in recent announcements about its own chips. In a blog on May 5, company spokesman Isaac Schultz noted:

“While Trainium focuses on AI training and inferencing, AWS Graviton processors handle the sustained workloads that keep the internet running and increasingly power the agentic AI era. Graviton delivers up to 40% better price performance than comparable x86 processors and uses less energy for the same output. More than 100,000 customers use Graviton-based servers today…. That's why Meta is deploying tens of millions of Graviton cores to power the CPU-intensive workloads behind agentic AI, with the performance and efficiency they need at their scale.”

During the company’s Q1 2026 earnings call, CEO Andy Jassy noted the importance of Graviton processors to the company’s plans: “AI is commonly seen as a GPU story, but the rise of agentic workloads, real-time reasoning, code generation, learning, and multi-step task orchestration is driving massive CPU demand as well.”

Amazon stated in its earnings presentation that Graviton CPUs are now used by 98% of the top 1,000 EC2 customers.

Google has its own take on CPU momentum. While the company offers its own Axion Arm-based CPUs as the basis for some AI/ML workloads and also uses Intel Xeon processors for some cloud instances, Google recently announced with Intel a multiyear agreement in which Google will work with Intel to develop ASIC-based infrastructure processing units (IPUs). These new programmable accelerator chips will, according to Google and Intel, “offload networking, storage and security functions from host CPUs - improving utilization, increasing efficiency and enabling more predictable performance across hyperscale AI environments.”

Pros and Cons

Enterprise customers will no doubt welcome the increased focus on CPUs as key to AI deployment, since CPUs are typically more energy efficient and cheaper than GPUs. As one Reddit poster stated:

“Market finally figured out that CPUs are just as important as GPU and TPU for AI because that’s where the actual code execution happens. You need the CPU for the logic, fetching data, and running cron jobs while the TPU or GPU just handles the math. Intel and AMD are ripping because people realized you can’t run ‘agentic’ AI without a massive amount of traditional CPU power to manage the workflow.”

A potential downside of the new CPU movement is the lag between announcement and delivery. All of the product rollouts announced this week, plus the increased total market figures, won’t materialize in full force until next year. That said, the direction seems clear: CPUs are on the menu as increasingly essential to inference and agentic AI.

Futuriom Take: Recent chip announcements highlight the expanding role of CPUs in AI workloads, as demand for inference and agentic AI require preprocessing of data and sequential computing. Despite the long-range outlook, this market is building momentum and showing that the future of AI will include more chip diversity.