iqshard · one server · 8 × NVIDIA B300
Your AI stops because the allowance ran out. Ours doesn’t have an allowance.
Rent a Share of a dedicated 8 × NVIDIA B300 server running the latest GLM. Flat monthly price, no metering, no weekly reset, no 11pm wall.
Server
GPU
8 × NVIDIA B300
Memory
8 × 288 GB HBM3e (2.304 TB)
Model
Latest GLM
What you’re renting.
We buy one server outright and split it into Shares. You rent a Share, month to month.
Point your coding agent at it and send whatever you want. Nothing counts anything.
When the server is busy your request waits a few seconds behind someone else’s. That is the only thing that ever happens to you.
From
$5pcmper Share
Pro Shares cost a little more and take first claim when the server is loaded. Both are flat. Neither has a cap. See pricing →
Why it works
A $5 Share stays $5.
One server costs what it costs whether one person or two hundred are on it, so the price of a Share is that bill divided. You join at $5 and you stay at $5, and that is the whole bill however hard you work it.
Everyone on the server gets a turn.
Priority classes set the order of the queue, and the queue keeps moving. On a busy server your turns come a little further apart. The model, the server and the day are the same ones you had when it was quiet.
The latest GLM, on hardware we own.
For everyday agentic coding this is the sweet spot: the quality gap to frontier models is far smaller than the price gap. We publish the comparisons so you can judge that for yourself, and the most recent one is below.
A server is open now.
While a server is funding you pledge, and nothing is charged until it is confirmed as proceeding. Once a server is confirmed or live you buy outright, and that is charged when you buy it. Either way your access starts when the server goes live.