ZETTA AI Box Hardware & Specs | 1 PFLOP On-Premise AI

NVIDIA GB10 Grace Blackwell

Data-center AI performance in a 150 × 150 mm footprint

ZETTA AI Box is built on the NVIDIA® GB10 Grace Blackwell Superchip, bringing enterprise-class AI to a compact, portable appliance.

1 PFLOP

AI compute comparable to entry-level enterprise AI servers

128 GB

Coherent unified GPU memory

200B

Model parameters: runs large models locally

150 × 150 × 51 mm

Fits in a backpack

Ports & connectivity

Six interfaces, one compact rear panel

ZETTA AI Box · rear panel

Power button

USB-C power-in

3× USB-C Gen 2x2 with DisplayPort Alt mode

HDMI 2.1b

10GbE RJ-45

NVIDIA ConnectX-7 Smart NIC, 2× 200G QSFP

Illustrative schematic. Port order and positions vary by manufacturer.

Full specifications

Technical specifications

Enterprise-class AI performance in a compact, portable appliance. Hardware configurations are based on the NVIDIA GB10 platform.

Dimensions, connectivity details and security features may vary by manufacturer. All ZETTA AI Box configurations include 4 TB storage. * ZETTA AI Box is also supported on other NVIDIA GB10 platforms with equivalent specifications, including ASUS and Lenovo.

Hardware platform*

Dell Pro Max GB10

Processor

NVIDIA GB10 Grace Blackwell Superchip

CPU

20 Arm cores (10× Cortex-X925 + 10× Cortex-A725)

GPU

NVIDIA Blackwell integrated GPU

AI performance

Up to 1 PFLOP FP4 (with sparsity)

Memory

128 GB coherent unified memory, 273 GB/s

Storage

4 TB NVMe SSD

Local AI models

Up to 200B parameters, depending on quantization

Networking

10GbE Ethernet, NVIDIA ConnectX-7 (2× 200 Gbps QSFP)

Wireless

Wi-Fi 7, Bluetooth

Connectivity

USB-C, HDMI, Ethernet, QSFP

Security

TPM-backed security; storage encryption options vary by platform

Dimensions

Approx. 150 × 150 × 51 mm

Weight

Approx. 1.2–1.5 kg

Operating system

NVIDIA DGX OS (Ubuntu Linux)

Extensibility

Up to 3 interconnected devices via QSFP, with up to 384 GB combined memory, subject to software support

Deployment modes

Standalone · Network Appliance · Access Point

Internet required

Extensible

Scale your AI capabilities

Connect multiple ZETTA AI Boxes for greater processing power. High-speed QSFP links and NVIDIA ConnectX-7 networking enable distributed AI workloads, larger AI models and expanded processing capacity. All processing remains on-premise, without cloud dependencies.