Mission control for AI infrastructure

Buy the hardware
the model actually needs.

From a single desktop card to a 72-GPU datacenter rack — Aiza's Assignment 9 Electronics Incorporated shows you the real specs, the real price, and a plain-English reason for every number, so you buy exactly enough hardware and not a dollar more.

Price range
$4,499 – $4.0M
Biggest single build draws
129 homes worth of power
Open-source models advised
3
The hardware line

Five tiers. Real specs. No guessing.

NVIDIA GeForce RTX 5090

Desktop card

Memory 32 GB
Price $4,499

NVIDIA H100 SXM5 80GB

Single datacenter GPU

Memory 80 GB
Price $35,000

NVIDIA H200 SXM 141GB

Single datacenter GPU (bigger memory)

Memory 141 GB
Price $38,000

NVIDIA HGX H200 8-GPU Server

Datacenter server

Memory 1,128 GB
Price $370,000

NVIDIA GB300 NVL72 Rack

Full datacenter rack

Memory 20,480 GB
Price $3,850,000
Model advisory

Three frontier open models. Real memory math.

Every recommendation starts with the same question: how much GPU memory does this model actually need? See the math on the Models page.

MIT

GLM-5.2

Z.ai · 744B parameters

Memory needed 893 GB
MIT

DeepSeek V4 (Pro)

DeepSeek · 1600B parameters

Memory needed 1,920 GB
Modified MIT

Kimi K2.7 (Code)

Moonshot · 1000B parameters

Memory needed 1,200 GB