AMD’s Venice Server CPU Demand Reportedly Extends Into 2028

AMD’s next-generation EPYC Venice server processors are reportedly seeing such strong demand that much of the company’s expected 2027 production has already been sold, with industry channel checks suggesting AMD is now taking orders extending into 2028.

The reports come as server CPUs take on a changing role in AI systems. While GPUs and dedicated AI accelerators perform the bulk of AI model training and inference, CPUs remain responsible for orchestrating workloads, handling data processing and storage operations, running virtualization and managing the broader software stack surrounding accelerators.

The reports on X by someone named Jukan (who has a pretty strong track record) point to unusually strong customer interest in AMD’s new Zen 6-based server platform.

Venice is a substantive upgrade for AMD’s EPYC server line. The processor family is being manufactured on TSMC’s N2, or 2nm, process technology, making it one of the first major high-performance computing products to use the node.

AMD has been positioning Venice for conventional enterprise and cloud computing as well as increasingly demanding AI infrastructure workloads. Intel and Nvidia are adopting similar strategies with their respective CPUs.

Venice is expected to span a wide range of configurations. AMD’s high-end SP7 platform is expected to scale to as many as 256 Zen 6c cores and 512 threads, while other versions will target different combinations of compute density, memory bandwidth and workload requirements. Another SP7 variant is Venice-X, which features up to 96 cores and 192 threads.

AMD began production of Venice on TSMC N2 in May and said it eventually plans to manufacture Venice at TSMC’s Arizona facility in addition to Taiwan. The 2nm node is important to AMD as it attempts to increase performance while maintaining or lowering power requirements.

One thing to note about Venice is that AMD isn’t just throwing cores at the problem. AMD is also pushing memory and I/O substantially harder. AMD claims Venice can reach roughly 1.6 TB/s of theoretical memory bandwidth, nearly three times the bandwidth of Turin, the previous generation.

AMD will begin shipping the SP7 version of Venice processors beginning this quarter. The low power SP8 processors will begin shipping in Q1 of 2027.