For neoclouds, service providers and CSPs
Zero
idle.
Sell every GPU 2.5 times.
Your tenants pay for a whole GPU and use about 20% of it. zeroidle™ turns the rest into capacity you can sell, and brings the customers to buy it.
Your hardware and your costs stay as they are. Proven over six months on packet·ai, our own neocloud.
One GPU, one month, at $5 an hour
1,825 billable hours from the same 730-hour month
Where the money is hiding
The GPU you sold is 80% empty.
A tenant rents a GPU from you and uses it about a fifth of the time. You've paid for every one of the other hours, in power, cooling and financing. Today you can't sell them to anyone else.
Training tenants use about 30% of a GPU. Inference averages 10-15%.
How zeroidle™ works · 1. The platform
One GPU, sold more than once.
The hosted·ai platform runs on the GPUs you already have. Several tenants share a GPU, and each still gets the whole card the moment they need it. On average, providers on zeroidle™ sell every GPU 2.5 times.
Each block is 5 hours a tenant rented. The dashed line is 730 hours, where a GPU sold once stops.
610 + 470 + 360 + 250 + 135 = 1,825 billable hours. Five tenants, one GPU, one month.
Proven in production
We proved the 2.5x on ourselves.
packet·ai is our own neocloud, and it runs on hosted·ai. Six months in, with paying tenants and hundreds of thousands of GPU hours behind it, the average GPU has been sold 2.5 times. That's where the number on this page comes from.
How zeroidle™ works · 2. The customers
Then we bring the buyers.
Capacity nobody buys is worth nothing, and most providers don't have a sales team for it. We do. Your spare capacity is sold through our own channels.
Our GPU marketplace. Your capacity goes in front of buyers who are already looking for GPUs.
packet·aiOur own neocloud. It owns no GPUs and sells capacity on our partners' infrastructure by the hour.
And underneath: Token Factory
It vacuums up whatever is left.
The last free tensor cores and the last megabytes of GPU memory run inference and get sold as tokens. Tenants always come first. It's the last step to zeroidle™.
One GPU, earning 2.5 times as much
We call it
zeroidle™.
It's what happens when you combine the software to share GPUs with the demand to keep them busy. Every GPU you own, earning every hour it can.
What it means for you
One GPU, before and after.
Your hardware stays yours.
zeroidle™ runs on hosted·ai software, on the GPUs you already own, in the data centre they're already in.
Your off-take contracts don't change.
Clusters sold to an offtaker stay sold 1:1. Nothing is shared unless you put it on-demand.
Your tenants keep the full GPU.
Each one gets the whole card the moment they need it. We run GPUs to 80-90%, not 100%, to keep that headroom.
Your numbers
Your fleet, with and without.
Nothing about the fleet changes except how often each GPU is sold. Move the sliders to match yours.
+$26.3M a year from GPUs you already own.
A simplified illustration. It leaves out Token Factory revenue and hosted·ai's fees. Payback is revenue against purchase price, before running costs.
You bring the capacity. We bring the cloud and the customers.
Tell us how many GPUs you run and how they're sold today. We'll show you what the same fleet earns with zeroidle™.