Lost in a downbeat day for tech, was a big moment for Advanced Micro Devices. During an event Thursday afternoon called Advancing AI, the company made a compelling case that it's becoming a true rival to Nvidia.
AMD's latest data-center confidence is showing up in projections for the future. Normally a company that is conservative with financial estimates and comparisons to competitors, AMD sounds suddenly ebullient about its prospects.
During Thursday's keynote, led by CEO Lisa Su, the company once again boosted its projections for the AI market. AMD also made bold claims about its chips' superiority versus Nvidia, Intel, and Arm Holdings, out-of-character bravado for the firm.
The announcements show just how far AMD has come in the past three years, from everyone's second choice to a leading chip vendor in the red-hot AI data center arena. In a world previously dominated by Intel and Nvidia, AMD is staking its claim to the hundreds of billions of dollars flowing from the AI boom.
When the AI frenzy began, the typical AI server had two Intel CPU chips and eight Nvidia GPUs, with four of the servers going into a standard-sized rack. AMD was largely shut out. The company had only competed with Nvidia in gaming chips, not in the data center.
By 2024 Nvidia had reimagined what a server could be by stuffing 72 of its GPUs, 36 of its CPUs, more Nvidia networking chips, and miles of copper cabling into a single server, the size of the whole rack.
AMD was coming from way behind. A year ago at Advancing AI, AMD announced its alternative, the Helios rack with AMD GPUs, CPUs, and networking chips.
During Thursday's keynote, AMD said Helios was cheaper, faster and more efficient than Nvidia's current Vera Rubin servers.
AMD also upgraded its projection for the total market for AI chips in 2030 to $1.4 trillion, up from the $1 trillion it shared at its analyst day in November.
Helios is in production and will be shipping to a list of customers that now includes Microsoft, Oracle, Meta Platforms, OpenAI, and Anthropic. The latter three participated in the keynote presentation -- a clear endorsement of AMD's progress.
Nvidia remains the first choice for AI servers, but the Helios rack looks competitive. It's likely to see plenty of business as customers seek to lessen their reliance on Nvidia.
The rise of agents, which use AI models to accomplish a complex series of tasks from simple conversational commands, is driving the booming market -- and AMD's new opportunities. The process of running AI models on these rack-servers is known as inference, and agents chew through it at a pace that no human can come close to matching. Inference-hungry agents are the primary reason that AMD can float a number like $1.4 trillion.
Agents are software, which require traditional CPU servers.
In November, AMD said that it thought that the total server CPU market would be about $60 billion in 2030. The company doubled that estimate in May. On Thursday AMD raised the bar again, now to $220 billion. That's a 267% increase in just eight months.
The server CPU market is one where AMD was already very competitive, in this case with former market leader Intel. In November, AMD claimed to have 40% of this market, with a clear path to 50%. The field is increasingly crowded, with data center CPUs also available from ARM, Nvidia, and Qualcomm.
Here too, AMD made bold claims of around the performance and efficiency of its new sixth generation of EPYC CPU processors.
Another emerging thread in the rise of inference computing is the need to separate different parts of the process for certain types of workloads that demand high speed and low latency. New chips built for that purpose are being paired with the rack-servers. Nvidia has its own solution through its $20 billion deal with chip start-up Groq.
On Thursday, AMD announced a partnership with the recently public chip company Cerebras that will perform this sort of "disaggregated inference" for these high-speed tasks.
AMD's announcements on Thursday weren't all new, but together they show a company producing an Nvidia-like approach to the AI data center -- one that includes multiple chips, different kinds of servers optimized for their roles, and software that ties it all together.
"We're building the broadest set of compute engines in the industry so that we have the right compute for the right workload," CEO Lisa Su said during the keynote.
Investors may not be buying in yet. AMD stock was down 2.7% after the keynote news while Cerebras shares were up 4.0%.
The message should be more clear: AMD has come a long way in three years.
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
July 23, 2026 14:45 ET (18:45 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments