HPE unveiled the ProLiant Compute DL394 Gen12, a 2U server built around NVIDIA Vera CPUs. Unveiled at Computex 2026, the machine targets agentic AI workloads, promising higher CPU throughput and lower latency than traditional x86 designs. Antonio Neri, HPE’s CEO, framed the launch as a response to the shift from generative to agentic models.

The server’s monolithic Vera chip avoids NUMA penalties, delivering 1.2 TB/s aggregate memory bandwidth—roughly 14 GB/s per core—via LPDDR5X modules. Integrated Lights-Out 7 and a silicon root‑of‑trust secure enclave meet NIST’s quantum‑resistant standards, while HPE Compute Ops Management offers a single pane of glass for AI‑driven operations.

NYSE plans to run the DL394 in its trading infrastructure, citing over 1.1 trillion daily messages and the need for ultra‑low latency. By pairing HPE hardware with Redpanda’s streaming platform, the exchange aims to double efficiency compared with legacy x86 stacks. The server ships in fall 2026 as part of HPE’s AI Computing portfolio.

For enterprises, the DL394 promises a turnkey solution for AI‑intensive tasks such as real‑time risk analysis and autonomous decision engines. Its unified management console reduces admin overhead, and the quantum‑ready security posture eases compliance burdens in regulated sectors. Availability this fall positions HPE to capture market share from vendors still relying on fragmented x86 clusters.