AMD's 6th Gen EPYC claims up to 2.2x over Intel's flagship Xeon

AMD says its top-end 6th Gen EPYC 9996 processor delivers up to 2.2 times the performance of Intel's Xeon 6980P and up to 3.7 times the performance of Nvidia's Vera CPU. The company announced the full 6th Gen EPYC server lineup and the new EPYC Embedded 9006 Series on November 11, 2026, and the server chips are already shipping in systems from major hardware vendors.

The launch is AMD's most direct attempt yet to claim a central role for CPUs in AI data centres at a moment when accelerators dominate spending and attention. The new range runs on AMD's Zen 6 architecture and targets enterprise applications, cloud computing, high-performance computing and AI workloads.

Why does the CPU matter in an AI data centre?

Chipmakers are still working out what the processor's job is in systems built around GPUs. AMD's answer: the CPU remains central to orchestration, data access, enterprise applications and host functions around AI models.

The 6th Gen EPYC portfolio is designed for distinct CPU roles in AI infrastructure, according to AMD. Those include general-purpose services, GPU host nodes and agent sandbox environments. Cloud deployments are planned across major providers, including Microsoft Azure and Oracle Cloud Infrastructure.

Dan McNamara, Senior Vice President and General Manager, Compute and Enterprise AI at AMD, framed the launch as a bet on x86 endurance. "AI is significantly expanding the role of the CPU across every layer of the data center," McNamara said. "With 6th Gen AMD EPYC, customers can standardize on the proven x86 foundation while choosing a workload-optimized solution to run more agents, keep accelerated infrastructure fed and power the business-critical applications they depend on."

What do the benchmark numbers say?

AMD's headline claims pit its flagship against both its traditional rival and the CPU Nvidia built for AI systems:

• EPYC 9996 delivers up to 2.2x the performance of Intel's Xeon 6980P
• EPYC 9996 delivers up to 3.7x the performance of Nvidia's Vera CPU
• The line offers up to roughly 3.3x the agents-per-watt of Nvidia Vera for agentic AI workloads

The agents-per-watt figure matters because energy efficiency has become a primary constraint on AI infrastructure buildouts, and agentic workloads are among the fastest-growing categories. AMD did not publish independent verification alongside the numbers, so buyers will test them against their own workloads.

Who is deploying the chips?

Server systems using the processors are shipping now from leading original equipment manufacturers and original design manufacturers, AMD said. Meta is working with AMD as a lead customer and has been validating platforms based on the new chips ahead of wider deployment.

Microsoft Azure is preparing several services tied to the new processors:

• Azure HDv2 for AI data processing, agentic AI and high-performance computing
• Azure HXv2 for electronic design automation and technical computing
• Azure ND MI455X v7, based on AMD Helios, for production-scale AI inference

Oracle Cloud Infrastructure is also set to use the processors in OCI Compute E7 instances for enterprise applications and CPU tasks surrounding AI models. AMD presented the breadth of support as evidence that the product line targets both conventional enterprise computing and newer AI-linked infrastructure.

What is the EPYC Embedded 9006 Series?

Alongside the server announcement, AMD introduced the EPYC Embedded 9006 Series. It targets semiconductor test equipment, network security and enterprise storage. The chips offer up to 256 cores, next-generation connectivity and seven-year product longevity, according to AMD.

The embedded release pushes the EPYC brand beyond mainstream server deployments. Processor vendors are trying to capture demand across a wider range of infrastructure categories, and buyers in embedded markets often prioritise long support cycles and stable supply as much as raw compute performance. A seven-year longevity commitment speaks directly to that purchasing calculus.

How does the Helion rack strategy work?

AMD also outlined its Helion rack approach for AI deployments. The designs include liquid-cooled and air-cooled configurations, plus a setup that combines CPU execution with local GPU inference in one rack. Some configurations can reach up to 73,728 threads in a single rack, according to AMD.

The pitch reflects a shift in how operators buy infrastructure. Interest is moving from standalone server performance toward rack-level design, as operators balance compute density, energy use, cooling and networking. Across the market, suppliers increasingly package CPUs, GPUs and interconnects as integrated infrastructure rather than separate components. AMD's Helion lineup positions the company to sell that whole package rather than compete chip by chip.

What comes next for EPYC?

AMD's launch extends a product line that has steadily gained share in the server market over several generations. The company said its 5th Gen EPYC family was the fastest-ramping and highest-volume EPYC generation in its history, setting the backdrop for this release.

For customers assessing server platforms, the decisive questions will centre on performance under real workloads, energy use, system availability and software support across existing estates. Systems powered by 6th Gen EPYC server processors are shipping from leading server vendors now, AMD said, and the Azure, Oracle and Meta deployments signal how quickly the chips reach production AI infrastructure in the months ahead.

By Elena Vasquez