HP ZGX Nano AI Station
Own YourIntelligence
A palm-sized supercomputer. 200B parameter models. No cloud. No rent. Forever yours.
1,000
TOPS
128GB
Unified Memory
The breakthroughWhy a palm-sized box outperforms
Why a palm-sized box outperforms
racks of servers.
The Old Way
CPU and GPU had separate rooms
CPU RAM
Bottleneck
GPU RAM
Every time the AI needed data from one to use in the other, it copied billions of numbers through a tiny pipe. Most of the time, the model was just waiting.
The ZGX Way
One giant shared room
+
128GB
Unified coherent memory
No pipe. No copying. No waiting. CPU and GPU share one giant pool. That's why a palm-sized box runs models that would choke a rack.
Physics, not magic.
That's the whole trick.
Built different
Every component engineered for AI.
Superchip
NVIDIA GB10
Grace Blackwell. CPU + GPU on one die.
Memory
128GB
Unified, coherent.
Storage
4TB NVMe
Self-encrypting.
Network
200Gbps
Dual QSFP. Pair two for 256GB.
Model capacity
200B+
Parameters at INT4 quantization.
LAN
10GbE
Direct cable to your network.
Posture
Air-gap
Zero egress capable.
Form factor
Palm-sized
150mm. 2.6 lb.
The math is simple
One purchase versus endless bills.
Cloud AI
Pay per token, forever
Input tokens$15–30 / 1M
Output tokens$60–75 / 1M
Fine-tuning$25+ / 1K examples
Monthly (moderate)$500–2,000
3-year cost
$1,800
Light
$18,000
Moderate
$180,000
Heavy
ZGX Nano
One-time purchase
Input tokens$0
Output tokens$0
Fine-tuning$0
Monthly (any usage)$0
$3,600
One time. Forever.
Break-even at moderate usage: ~2 months
Supercompute.
Palm-sized.
ZGX Nano is shipping. Quantum Pipes is ready for it. Run frontier AI locally. Privately. Forever.
Air-gappedSelf-encrypting1000 TOPSCapsule governance