PopAI, local AI compute appliances
PopAI V128 adds real GPU capacity to a computer you already own, no rack, no cloud dependency, no new server to provision.
Your computer has discovered something.
Why PopAI
A conventional GPU server means a second CPU, a second bank of RAM, a second boot disk, a whole machine bought just to get four accelerators. PopAI assumes you already own a workstation and adds only the part you're missing: memory, compute, and an NVLink fabric.
If you need a whole server, buy a server. PopAI is for teams that already have the computer and need the accelerator half of an NVLink machine next to it, for private document and knowledge assistants, high-volume document processing, coding assistance, internal agents, research, and sustained local inference. You bring the workload, PopAI brings the capacity.
Qualified datacenter hardware
PopAI V128 is built on NVIDIA V100 SXM2 modules reclaimed from decommissioned datacenter systems, not new silicon. Every module is inspected, integrated, and load-tested as a four-way NVLink unit before it ships, with vonArgo support behind the finished appliance.
If your workload genuinely needs newer hardware, we'll tell you plainly when we talk through it.
Technical reference
| Compute | |
|---|---|
| Accelerators | 4× NVIDIA V100 SXM2 |
| Memory | 128GB aggregate HBM2, 900GB/s per GPU |
| Interconnect | NVLink 2.0, four V100s |
| Enclosure | ~630 × 330 × 140mm target, prototype specification pending |
| Cooling | Closed liquid loop, copper SXM2 cold plates, rear-mounted radiator and pump |
| Software | Arrives ready to run, no manual setup required |
Full software compatibility notes, benchmarks, and architecture detail move to a dedicated technical datasheet as PopAI leaves pilot. Pricing is set individually per unit based on configuration and lead time.
Pilot program
Tell us the workload you're running or planning, and what's already in place. We'll do a straight fit check, most conversations end with either a configuration and quote or an honest not yet, and why.
An honest read on whether PopAI is the right generation and scale for your workload.
Memory, host interface, and support terms scoped to your setup.
Assembly and extended load testing, with a build record for your unit.
Shipped and documented, with a support channel back to the people who built it.
Pilot units · limited availability · commercial terms supplied with quote