AMD splits Zen 7 into three EPYC families for 2028, extends roadmap to 2030

AMD held its Advancing AI 2026 event in San Francisco, launching sixth-generation EPYC "Venice" processors, Instinct MI400 Series GPUs, and Helios rack-scale systems. The company also confirmed that the upcoming Zen 7 generation, arriving in 2028, will launch as three separate EPYC families rather than a single lineup.

Three Zen 7 CPU Families Named

AMD named the three Zen 7 families in its launch release: Florence, Ferrara, and Fidenza. The company extended its annual CPU, GPU, networking, and rack cadence out to 2030 and projected its total addressable market at roughly $2 trillion by 2030.

AMD also introduced a new competitive metric, claiming the most AI agents per watt, per dollar, and per rack. However, the company's own endnotes state that agent counts are estimated from CPU thread resources used as a proxy.

Family Details

• Florence: Features fresh Zen 7 cores, a new set of AI compute extensions, and support for newer memory technologies, according to AMD Chair and CEO Lisa Su.
• Ferrara: Serves as the AI host node portion and reappears later in the roadmap as the CPU inside the Helios 600 rack alongside MI600 Series GPUs and Pensando "Palma" and "Levanzo" networking.
• Fidenza: Described as the agentic sandbox product.

AMD disclosed no core counts, no process node, and no socket for any of the three families, stating only that Zen 7 uses "leading-edge process technology."

Generational Shift in EPYC Strategy

The fourth-generation EPYC (Zen 4) spanned Genoa, Bergamo, Genoa-X, and Siena across two sockets. The fifth-generation "Turin" (Zen 5) consolidated Zen 5 and Zen 5c parts into a single 27-SKU stack on one socket with no separate cache-stacked or edge line at launch.

Venice restarts the fan-out approach:

• 9006 series on the new SP7 socket (now available)
• Venice-X arriving in 2027 with 1,152MB of 3D V-Cache, 96 cores, and a 5.15 GHz boost clock

Three named Zen 7 families announced two years out represents a wider generational spread than AMD has ever opened with.

Portfolio Segmentation

AMD's portfolio endnote describes the EPYC range as covering:

• General-purpose enterprise
• Cloud
• Telecom
• SMB
• HPC systems
• Sandboxed agentic AI deployments
• GPU head node servers (distinct category)

Su told analysts in May that AMD was already working with customers on architectures beyond Venice, without naming categories at the time. The Zen 7 lineup now puts names to those two AI-specific segments.

Agents Per Rack: Competitive Claims and Caveats

AMD's main server CPU claim is that sixth-gen EPYC enables the most agents per watt, per dollar, and per rack.

Endnote 9xx6-012 states that agent counts are estimates derived from available CPU thread resources used as a proxy under a consistent theoretical workload, and that real capacity varies with workload, model, memory, software, orchestration, and system configuration.

Per-Rack Comparison

Based on core count at a 100 kW rack power envelope:

• 256-core EPYC 9996 (AMD)
• 88-core Nvidia Vera
• 192-core EPYC 9965 (AMD)
• 128-core Xeon 6980P (Intel)

Per-Dollar Metric

Based on top-of-stack thread count divided by 1,000-unit list pricing.

Per-Watt Comparison (from endnote)

| Processor | Power | |-----------|-------| | Nvidia Vera | 450W | | Arm AGI CPU | 300W (1 thread/core) | | Intel Xeon 6980P | 500W | | AMD EPYC 9965 | 500W |

AMD had already claimed a 3.3x rack-level advantage over Vera in June.

Market Context

• Mercury Research: AMD at a record 46.2% of x86 server CPU revenue in Q1 2026 (vs. 33.2% of units)
• Arm-based designs: ~17.7% of server shipments in the same quarter

New Power Metric: Default CPU Power

Starting with sixth-gen EPYC, AMD has replaced TDP with Default CPU Power, defined as total power consumed across the processor's compute and I/O dies at a stated performance target. AMD says both references can serve for product comparison and performance-per-watt analysis. The endnote itself mixes conventions, quoting the EPYC 9956 at 400W Default CPU Power against TDP figures for Nvidia, Intel, and Arm parts.

Helios Rack Systems and 2030 Roadmap

Helios Rack Specifications (In Production Now)

• 72 Instinct MI455X GPUs
• 18 Venice CPUs
• 31TB of HBM4
• 1.4 PB/s aggregate memory bandwidth

AMD claims up to 30% more inference tokens per dollar than Nvidia's Vera Rubin NVL72, based on AMD Performance Labs estimates from July 2026 using a Kimi K2 Thinking workload at 32K input and 8K output, with hourly GPU pricing projections.

The 34x token throughput gain AMD quotes for MI455X over MI355X comes from AMD's own measurements on DeepSeek V4 Flash at FP4. Both figures are vendor-provided benchmarks with no independent verification yet.

Forward Roadmap

• 2027: MI500 Series GPUs inside a Helios 500 rack built on EPYC "Verano" and Pensando "Como" and "Monza" networking
• 2028: MI600 Series GPUs