ASUS has unveiled the XA P2N-E2, a 2U server with two NVIDIA Vera CPUs built for the execution layer of agentic AI: running code, using tools, isolating processes in sandboxed environments, processing data, and orchestrating tasks around the model. The company announced it on October 12, 2026, at the OCP Global Summit in San Jose, California, where it is also showing NVIDIA Vera Rubin NVL72 racks and servers with Intel Xeon 6 and sixth-generation AMD EPYC. The announcement includes no pricing or a specific availability date.
ASUS XA P2N-E2 with NVIDIA Vera key takeaways in 30 seconds
- 2U NVIDIA MGX server with two Vera CPUs, each with 88 Olympus cores.
- Up to 1.5 TB of LPDDR5X memory across 16 SOCAMM2 modules, plus two NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs.
- Aimed at AI agents, reinforcement learning, and data processing.
- Air-cooled only, with PCIe Gen6 slots and support for eight E1.S drives.
- ASUS has not given a price or availability date.
The server arrives amid a broader push to put the CPU back at the center of AI infrastructure. AI agents don’t just generate text: they compile, test, query databases, and run tools in a loop, and much of that work falls on the processor and memory rather than the GPU. That’s the argument NVIDIA has made with Vera, which we covered when explaining why the CPU is returning to the center of agentic AI.
Two Vera Chips, 1.5 TB of Memory, and Optional GPUs in 2U
According to ASUS’s spec sheet, the XA P2N-E2 packs two Vera processors with 88 cores and 176 threads each, linked by second-generation NVLink-C2C with up to 1.8 TB/s of coherent CPU-to-CPU bandwidth. Memory is based on the SOCAMM2 format, removable LPDDR5X modules designed for servers, and ASUS says it doubles the power efficiency of traditional RDIMMs. The chassis supports two dual-slot graphics cards and features CPU and GPU trays that can be serviced independently.
| Form factor | 2U, NVIDIA MGX architecture, air-cooled |
| CPU | 2 x NVIDIA Vera (88 cores / 176 threads each) |
| CPU-to-CPU interconnect | 2nd-generation NVLink-C2C, up to 1.8 TB/s |
| Memory | 16 x SOCAMM2, LPDDR5X at 9,600 MT/s, up to 1.5 TB |
| GPU | Up to 2 x NVIDIA RTX PRO 6000 Blackwell Server Edition (dual-slot) |
| Storage | 8 hot-swap E1.S bays (PCIe Gen6) |
| Expansion | 2 PCIe Gen6 x16 slots for NICs plus 1 dual-width slot for a DPU/NIC |
| Power | 2+2 redundant 2,000 W 80 PLUS Titanium supplies |
| Dimensions and weight | 900 x 438 x 88 mm; 40 kg net |
On performance, ASUS credits Vera with up to 1.8 times more performance per thread than traditional x86 CPUs and up to 1.8 times faster sandboxed code execution. These are vendor-supplied figures, with no detail on the configurations compared. The first benchmarks published point in the same direction for specific workloads, including Vera’s performance against AMD EPYC Zen 5 and its DuckDB result against Xeon 6, but it’s worth waiting for more independent measurements.
The Rest of ASUS’s OCP Showcase
The XA P2N-E2 is part of a broader lineup. At rack scale, ASUS is showing the XA VR721-E3, its AI POD based on NVIDIA Vera Rubin NVL72, with 36 Vera CPUs and 72 liquid-cooled Rubin GPUs. For more flexible configurations, it is presenting the XR R1I12B-M1BHFS, which combines NVIDIA HGX Rubin NVL8 with direct liquid cooling on the GPUs and air-cooled Intel Xeon 6 processors. Its booth also features the Spectrum-6 Ethernet network, the NVIDIA CMX context-memory platform, and the NVIDIA Groq 3 LPX inference accelerator.
Outside the NVIDIA ecosystem, ASUS is showing the RS700-E12-RS4U and RS720-E12-RS12U with Intel Xeon 6, the RS700A-E14B-R16U with AMD EPYC 9006 Venice, PCIe 6.0, and up to 16 E3.S drives, the ESC8000A-E13P with AMD Instinct MI350P accelerators, and the XA AM3A-E13, which packs eight AMD Instinct MI355X units into a 10U air-cooled chassis.
A CPU Already Reaching the Big Server Makers
ASUS isn’t the first to bring Vera to its own server. In June, Dell and HPE announced their own systems built around this CPU for enterprise agentic AI. What sets ASUS’s design apart is its form factor: 2U, air-cooled, and built on the MGX standard, which makes it easier to install in data centers without water loops. In exchange, GPU configuration is limited to two PCIe cards, far from full rack-scale systems.
On availability, ASUS says only that its servers are sold worldwide, that products and configurations may vary by region, and that customers should check with their local representative. Neither the announcement nor the spec sheet specifies when the XA P2N-E2 can be ordered or at what price.
Frequently Asked Questions
What is the ASUS XA P2N-E2?
It’s a 2U server built on NVIDIA MGX architecture with two NVIDIA Vera CPUs, designed for agentic AI, reinforcement learning, and data processing workloads.
How much memory does the XA P2N-E2 support?
Up to 1.5 TB of LPDDR5X at 9,600 MT/s across 16 SOCAMM2 modules, according to ASUS’s spec sheet.
What GPUs can it carry?
Up to two dual-slot NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs.
When will it be available and how much will it cost?
ASUS hasn’t published a price or date. It points buyers to its local representatives and notes that availability may vary by region.
Sources:
