Skip to content
Kinesis

Kinesis Batches

A fleet of GPUs,
made a breeze.

GPUs by the hour for rendering, AI and mining. Your Kinesis Grid keeps the work running as GPUs come and go.

Launch NowTalk to our team

Your Kinesis Grid

53 GPU models65 online now130 online this week

Each square is a GPU that was online in the last 7 days. They come and go through the day, and your batch spreads across whichever are online.

From $0.03 per GPU-hourLive · updated Nov 10, 3:54 PM UTCBuild your grid

Simple hourly pricing.

Every price below is a cap: the most you'll pay per GPU-hour. No bidding, no surprises. Add GPUs to build your grid and see the most your run can cost.

Live · updated Nov 10, 3:54 PM UTC

Individual GPUs come and go through the day, and your Kinesis Grid replaces any that leave with the next to come online. Add any GPU to your grid: it joins as soon as a GPU of that model is free.

Up to, per machine-hour

  • B3008 × 288 GBJoins when one is free$55.52/hr
  • B2008 × 180 GBJoins when one is free$53.92/hr
  • H2008 × 141 GBJoins when one is free$24.48/hr
  • H1008 × 80 GBJoins when one is free$21.60/hr
  • A1008 × 80 GBJoins when one is free$7.12/hr
  • L40S8 × 48 GBJoins when one is free$6.40/hr

Add GPUs from the list and your grid takes shape here. Each GPU runs on its own machine, and the totals show the most your run can cost.

Per GPU, per hour. Listed prices are caps: you never pay more than this. Made for work that runs in batches. Prices as of November 10, 2026.

Need more GPUs?
Let's size it together.

Dozens of GPUs, a mix of models or a long run? Tell us what you're working on and our team will get back to you with what's available and a price.

Prefer email? sales@kinesis.network

Takes two minutes. We'll match you with the right hardware.

Tell us about your run

Big queues.
Small bills.

If your work runs in batches, pay only for the hours it runs.

AI inference

Clear the backlog overnight.

Spread offline inference, embeddings and evals across many GPUs at once and get through the queue faster.

Batch inference · Embeddings · Evals
Rendering

Render more frames for less.

Send every frame to its own GPU. Animations, product shots and VFX finish sooner, and your budget goes further.

Animation · Product shots · VFX
Training & fine-tuning

Experiment more. Spend less.

Run LoRA fine-tunes and hyperparameter sweeps, one experiment per GPU, for the same budget. Save checkpoints to your own storage and pick up where you left off.

Fine-tuning · LoRA · Experiments
Need 24/7?

Running a website, database or live API that has to stay up around the clock? Kinesis runs those too, on dedicated compute with your private Grid included, from small CPU machines to 8-GPU H100 nodes. Spot is built for work that runs in batches. See dedicated compute →

From sign-up to running in minutes.

01

Pick your GPUs

Choose models and see exactly what they cost per hour, before you sign up.

02

Sign up and deploy

Create an account with Google, GitHub or email, add a card and launch.

03

Run and keep every result

Results and checkpoints go straight to your own storage, so every GPU-hour counts.

Good to know.

Everything you need before your first run.

Talk to our team

Capacity that would otherwise sit idle. You pay by the hour, never more than the model's listed cap. Any single machine can leave without notice when its owner needs it, but your Kinesis Grid replaces it with the next GPU that comes online, so your batch keeps running on one steady fleet.

Your next batch, made a breeze by Kinesis Grids.

Sign up with Google, GitHub or email and deploy to spot GPUs in minutes. Every hourly price is a cap.

Launch NowSee dedicated compute