By Adam Levine
It has been a sad few years for the venerable central processing unit. The one-time standard-bearer of the computing revolution has been pushed aside in recent years with the rise of generative artificial intelligence powered by graphics processing units, or GPUs, which are used for a lot more than graphics these days.
But now attention is swinging back to the main brain of data-center computing, the CPU. With one big change: Unlike in the past, Intel will have to share the spotlight, and may not even be the primary winner.
In the past two weeks, three server CPU companies -- Intel, Advanced Micro Devices, and Arm Holdings -- reported their earnings, and each gave commentary indicating that the market will be larger in a few years than they had previously thought.
"In recent months, we have seen clear signs that the CPU is reasserting itself as the indispensable foundation of the AI era," said Intel CEO Lip-Bu Tan in April. "This is not just our wishful thinking, it is what we hear from our customers."
Arm has projected that the server CPU market will hit $100 billion by 2030, and this week AMD doubled its own estimate to $120 billion. The three companies had $34 billion in combined data-center revenue last year.
That sort of market expansion would be a huge boon to these companies, but to keep it in context, Nvidia is expected to earn more than $150 billion in data-center sales in just the next two quarters.
Before the release of ChatGPT in 2022, servers rarely needed GPUs to do heavy lifting for intense mathematics. Expensive GPU-filled servers were overkill, and data-center customers mostly bought CPU servers to run websites, databases, applications, and other traditional workloads. A typical server configuration contained two Intel server CPUs.
In a 2021 two-way data-center sales race between Intel and Nvidia, Intel held a 70% market share. But the AI revolution changed that dramatically and by the end of last year Intel only garnered a 7% share.
But four years into the gen AI boom, CPUs are making a comeback. The reason is that AI workloads are beginning to shift from math-intensive training of AI models to the more varied process of running these models, known as inference. The big 2026 inflection is AI agents, software that uses AI models to complete a complex series of tasks from simple conversational prompts. If predictions turn out to be true, in a few years agents will proliferate to the point where they outnumber people on enterprise networks.
Agents chew through inference at a rate that no human could match, and so the hundreds of billions of dollars worth of data centers coming online this year are being built to support that trend. GPUs will still be the primary expense because of AI's intense math demands.
But agents are software, and software runs on CPUs. If it were five years ago, this would be unquestionably good news for one company: Intel. Indeed, in 2020 and 2021 there was a burst of Covid lockdown-fueled demand for CPU servers and Intel had its best years in the data center, with $26 billion in sales both years.
But the data center CPU market is far more complex now. For starters, server CPUs from Intel's longtime competitor, AMD, have been gaining in popularity and are considered by many to be better now. At its November investor day, AMD said it had around a 40% market share, and predicted that it will keep rising.
The bigger source of competition may come from chips that are based around the Arm CPU instruction set. An instruction set is the most basic level at which hardware and software communicate. The x86 instruction set that underpins Intel and AMD chips has been dominant in servers, but that's been changing.
Arm's first step into x86 server territory was the 2018 launch of the Amazon Web Services Graviton CPU. AWS is the oldest and largest of the clouds, and this set the tone for the industry. Graviton is on its fifth generation now, and for three years running, over half of AWS's new CPU capacity has been from Graviton. Now the other two big cloud players -- Microsoft Azure and Google Cloud -- also rent out servers with their own custom Arm CPUs. The advantage over x86 servers is Arm's power efficiency that allows for more computing in the same size data center.
Apple customers have firsthand experience with the efficiency of Arm's designs, which have been used in MacBook chips since 2020 and easily beat x86 chips in terms of speed and battery life.
This year, Arm changed the game, announcing it would sell a server chip, the AGI CPU, competing with its own licensees for the first time. The company recently predicted that it could see $15 billion in fiscal year 2031 chip sales and this week said near-term demand is twice what it expected.
Another shift toward Arm-based CPUs in the data center comes from the accelerated servers themselves. At first, Nvidia's AI servers sported Intel CPUs to orchestrate all the mathematics happening in the GPUs. But then in 2023 Nvidia switched over to using its own Arm-based CPUs, and others are now following. Google's latest TPU-accelerated server will feature its own Axion CPU, and Meta Platform's coming custom AI accelerator servers will be run by Arm AGI CPUs. With a nod to the power efficiency of these CPUs, Arm is becoming the choice for this data center workload.
AMD's $120 billion prediction should be considered an educated guess, but if the server CPU market expands that quickly, the competition will be fierce and there may not be room for Intel, AMD and Arm all to win.
Write to Adam Levine at adam.levine@barrons.com
This content was created by Barron's, which is operated by Dow Jones & Co. Barron's is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
May 08, 2026 16:34 ET (20:34 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments