AMD announced on Nov. 10 that its 6th Gen EPYC SP7 processors, built on the Zen 6 architecture and part of a family largely code-named "Venice," are now generally available. The company said in its press release that systems are shipping from leading server vendors and that instances on Microsoft Azure and Oracle Cloud Infrastructure (OCI) will be generally available soon.

The EPYC 9996 is the 256-core part AMD uses for its headline comparisons. Alongside the chip launch, AMD said the first of its Helion rack configurations, which were shown in October, are due from partners this month. For data center operators the two announcements are linked: Helion is AMD's rack-level design for agentic AI workloads built around these CPUs, and it is a separate product from the company's Helios racks.

What AMD Is Shipping

SP7 is the first of four 6th Gen EPYC parts, according to Data Center Knowledge. The other three are due in 2027. SP7 supports up to 256 cores.

The chips are already in customers' hands in some form. Data Center Knowledge reported that AMD CEO Lisa Su said on the company's third-quarter earnings call that Venice revenue shipments began in Q3. She also said hardware makers had committed to more than 240 Venice platforms, 20% more than for the previous generation, "Turin." The same report listed Dell Technologies, Hewlett Packard Enterprise, Lenovo, Supermicro and Cisco as hardware partners for Helion racks. AMD's release also says Meta is a lead customer validating 6th Gen EPYC platforms ahead of deployment at scale.

On power, footnotes in AMD's press release list a default CPU power of 600 W for the 256-core EPYC 9996. That figure comes from AMD's footnotes, not from a published specification sheet. Rack power and cooling planners should treat it as the company's stated default and not as a measured draw for any particular system. One footnote also describes an AMD analysis of 9996-based servers in a 35 kW rack, with a maximum of 20 servers per rack.

The Performance Claims, and Their Source

AMD makes several head-to-head claims for the EPYC 9996. The company says an EPYC 9996-based system delivers:

  • up to 3.7 times the performance of a system built on Nvidia's Vera CPU;
  • up to 2.2 times the performance of a system built on Intel's Xeon 6980P;
  • up to 2.0 times the performance per dollar of that Intel Xeon 6980P system.

All of these figures come from AMD's own benchmarks and are footnoted in its press release. The sources reviewed for this article contain no independent third-party testing of them. Results that start with "up to" describe the best case in AMD's chosen tests and will not necessarily hold for other workloads, configurations or software stacks. Buyers usually wait for vendor-neutral results and their own testing before relying on comparisons like these.

Dan McNamara, AMD's senior vice president and general manager of Compute and Enterprise AI, is quoted in the release on the launch.

Helion: Rack Designs for Agentic AI

AMD first showed Helion at the OCP Summit 2026 in San Jose in October. The Register reported at the time that Helion is a rack reference design aimed at AI agent workloads. It said partner-built configurations were scheduled for November, offered as CPU-only or mixed CPU/GPU systems, in liquid-cooled or air-cooled versions, and that Helion is separate from AMD's Helios racks.

AMD's Nov. 10 release names three options:

  • Helion Liquid, which AMD describes as designed to sit side by side with Helios racks in large-scale agentic AI deployments;
  • Helion Air, for expanding agent capacity within existing air-cooled data centers;
  • Helion Air Inference, which combines CPU execution and local GPU inference in one rack.

Helion Air may matter most to facilities that have not been converted to liquid cooling. AMD's own positioning of it for existing air-cooled data centers suggests it is meant for buyers who want to add CPU capacity for agent workloads without retrofitting cooling. Helion Liquid, by AMD's description, is intended to pair with Helios deployments.

Data Center Knowledge reported that AMD said the first Helion racks will become available throughout this month.

A Day Earlier: EPYC Inside Google's TPU Systems

On Nov. 9, the day before the Venice announcement, AMD announced that 5th Gen EPYC processors power the host infrastructure for Google Cloud TPU 8tx. AMD described TPU 8tx as the first cloud TPU system with hardware-enforced confidential computing, which uses AMD's Secure Encrypted Virtualization (SEV) technology. The release includes statements from McNamara and from Mark Lohmeyer of Google Cloud.

The two announcements make the same case from different directions. In large accelerator deployments, including Google's TPUs as well as GPUs, AMD says EPYC CPUs handle host processing, such as preparing data and coordinating workloads. AMD is selling EPYC both as that host processor and as the main compute engine in its Helion racks.

Why It Matters for Facilities

The launch has practical consequences for people who plan and run data center capacity:

  • Procurement timing. According to AMD, Venice systems from leading server vendors are shipping now. Cloud capacity on Azure and OCI is described only as coming "soon," although AMD says Azure HDv2 is already in preview.
  • Power density. The 600 W default CPU power in AMD's footnotes for its 256-core part matters for per-rack power budgets, especially in air-cooled rooms. Actual system-level figures will depend on vendor configurations.
  • Cooling choices. Helion is offered in both liquid and air versions, so operators can choose between deploying in existing air-cooled facilities and pairing with Helios racks.
  • Platform breadth. Su's figure of more than 240 committed Venice platforms, as reported by Data Center Knowledge, suggests buyers will have a range of system designs to choose from.

What Happens Next

According to the sources, the next steps are:

  • November 2026: partners deliver the first Helion racks. Data Center Knowledge reports availability throughout this month, and The Register reported in October that partner-built configurations were scheduled for November.
  • "Soon": Azure and OCI instances built on 6th Gen EPYC SP7 reach general availability. AMD has not given a date.
  • 2027: the remaining three of the four 6th Gen EPYC parts arrive, according to Data Center Knowledge.

Sources