The chip makers are best known for competing in the GPU business, but they're stepping up their offerings in the attractive market for server CPUs
AMD hosted its Advancing AI event on Thursday.
Advanced Micro Devices has been coming after Nvidia in the market for graphics processing units. But the smaller chip company's new product announcements show there's fierce competition in the market for central processing units as well.
AMD $(AMD)$ on Thursday unveiled the sixth generation of its EPYC central processing units for data-center servers at its Advancing AI event. The Venice family of chips is meant to support agentic AI workloads, AMD CEO Lisa Su said during a keynote presentation.
With the highest thread density, the chips can support more AI agents per watt, per dollar and per chip rack, according to the company. High thread density refers to the ability to run massive amounts of independent tasks at the same time on a single chip.
The launch comes after Nvidia (NVDA) shared performance data for its first standalone AI CPU, known as Vera, earlier this week. Nvidia introduced Vera in March and said it opens up a "brand new" total addressable market for the company, which has so far been the leader in the market for GPUs.
Vera is designed to maximize single-thread performance, which refers to a chip's ability to execute a single chain of instructions. Compared to AMD's fifth-generation Turin CPUs, Nvidia said Vera delivers up to 1.9 times faster agentic performance and six times better latency.
Even before AMD's latest announcements, Bank of America analyst Vivek Arya said he saw an "emerging battle" in the server-CPU market as the two chip makers look to cash in on the agentic AI boom.
High thread density and single-thread performance measure two different aspects of compute. AMD's approach focuses on throughput, or the number of tasks that can run at once, while Nvidia's technique targets latency, or how quickly a task is done.
Therefore, "the key question for investors is whether agentic AI is primarily constrained by time to complete an agent or number of agents per rack," Arya said in a Wednesday note.
Additionally, AMD's EPYC processors are built on the x86 instruction set architecture pioneered by Intel $(INTC)$, while Nvidia's Vera CPUs are based on Arm's energy-efficient architecture, setting up "a parallel debate" for investors, Arya noted. Going forward, he said x86's legacy "could become an important differentiator," as more software applications can run on that architecture.
AMD CEO Su said Venice CPUs are in full production and are seeing the strongest-ever demand from customers for any new generation of EPYC processor. She also announced that the next-generation Florence CPUs are expected to launch in 2028, followed by the eighth-generation Ravenna in 2030.
AMD expects the market for server CPUs grow to more than $200 billion by 2030, Su said. Last quarter, AMD reached a record 46% revenue share in the server-CPU market, according to Su.
Don't miss: Why cheap Chinese AI models could actually be a boon for Nvidia, Micron and other chip stocks
The chip maker also announced that its first-ever rack-scale solution, Helios, is in full production for multigigawatt-scale deployment by customers. Shipments are on track to start at the end of third quarter, Su said, and will ramp going into fourth quarter. She added that AMD plans to release a new Helios system every year.
AMD's stock fell 2.3% in Thursday's trading session, but were up 1.7% in after-hours trading, at last check.
-Britney Nguyen
This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
July 23, 2026 17:59 ET (21:59 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments