Skip to content
Kinesis

The compute we supply

Dedicated compute.
Your private Grid included.

Get the compute your applications need, supplied through Kinesis and operated as your own private Grid. Run workloads, share capacity across approved teams, and add more with one consistent way to deploy and manage it.

Dedicated hardware · All-in hourly or monthly pricing, Grid included · One way to deploy

The smaller question

When what you own is not enough.

Sometimes a deadline lands on a fleet that cannot grow. Sometimes a company needs a kind of GPU it cannot get anywhere soon. Sometimes a team wants to watch its application run this afternoon and has no patience for a purchase order. Our compute covers all three, and the point is how little a customer has to add rather than how much.

Start on ours. Nothing to buy first.

A team starts on our machines. Later it brings its own hardware into the same Grid, and its applications never notice.

Bought already. Need somewhere to go at the peaks.

The money is spent and the plan is to run on what was bought. That Grid runs mostly on its own machines and reaches into ours when the work spikes.

Two clouds and a data center. One screen.

Three environments, three contracts, three ways of running. Inside a Grid they are one system with one way to deploy.

Same product every time. What is inside the Grid changes. The way a team deploys does not.

Rates

What Kinesis-supplied compute costs.

Rates for the machines we supply, quoted per GPU per hour or per vCPU per hour, with your private Grid included. NVLink and SXM systems are sold as eight-GPU nodes. Rates are effective July 2026. For servers you already own, the Grid fee is on the pricing page.

A100$1.35

Unit Per GPU / Per HourGrid included

Specs 1x GPU, 28 CPUs, 120GB RAM, 750GB Storage

Notes Available in 1x, 2x, 4x GPU configurations.

H100$2.50

Unit Per GPU / Per HourGrid included

Specs 1x GPU, 28 CPUs, 180GB RAM, 750GB Storage

Notes Available in 1x, 2x, 4x GPU configurations.

A100 NVLink$1.50

Unit Per GPU / Per HourGrid included

Specs 8x GPU node: 252 CPUs, 1920GB RAM, 6500GB Storage

Notes Priced per GPU, sold as 8-GPU NVLink nodes.

H100 NVLink$2.75

Unit Per GPU / Per HourGrid included

Specs 8x GPU node: 252 CPUs, 1440GB RAM, 6500GB Storage

Notes Priced per GPU, sold as 8-GPU NVLink nodes.

H200 SXM$4.25

Unit Per GPU / Per HourGrid included

Specs 8x GPU node: 176 CPUs, 1800GB RAM, 48000GB Storage

Notes Priced per GPU, sold as 8-GPU NVLink nodes.

B200 SXM$6.50

Unit Per GPU / Per HourGrid included

Specs 8x GPU node: 252 CPUs, 2048GB RAM, 40000GB Storage

Notes Priced per GPU, sold as 8-GPU NVLink nodes.

B300 SXMfrom $8.13

Unit Per GPU / Per HourGrid included

Specs 8x GPU node: 252 CPUs, 2048GB+ RAM, 40000GB+ Storage

Notes Spot capacity. Price varies with market conditions.Check availability

Compute Optimized CPU$0.035

Unit Per vCPU / Per HourGrid included

Specs 1 vCPU, 2GB RAM, 50GB NVMe

Notes Available in 2x, 4x, 8x, 16x, 32x and 64x configurations.

General Purpose CPU$0.045

Unit Per vCPU / Per HourGrid included

Specs 1 vCPU, 4GB RAM, 50GB NVMe

Notes Available in 2x, 4x, 8x, 16x, 32x and 64x configurations.

Memory Optimized CPU$0.055

Unit Per vCPU / Per HourGrid included

Specs 1 vCPU, 8GB RAM, 50GB NVMe

Notes Available in 2x, 4x, 8x, 16x, 32x and 64x configurations.

Prices shown are for representative configurations. Actual specifications may vary by provider. GPU rates are per GPU per hour. NVLink and SXM systems are sold as 8-GPU nodes. Rates marked “from” are spot rates and vary with market conditions.

What we supply

Vetted machines, dedicated to you.

Kinesis provides dedicated compute in vetted datacenters and Kinesis-controlled environments, reserved for your organization or a group you choose. You control who shares your hardware. Kinesis manages reliability as part of the platform, from workload placement to recovery when infrastructure fails.

GPU compute

Capacity for training, fine-tuning, inference, and GPU experimentation. Choose by model requirements, GPU memory, GPU count, and interconnect where it matters.

CPU compute

Capacity for services, databases, and batch work. Choose by cores, system memory, storage, and the demand you expect.

Built for demanding workloads
AI/ML Training
AI/ML Inference
HPC & Simulation
Analytics & Query
Media Processing
Web & App Infrastructure

We secure capacity through direct agreements with datacenter operators and infrastructure owners, then bring it into one managed compute platform. We expand our supply around the hardware, locations, and scale our customers need. Capacity is distributed across multiple facilities and operators to reduce dependence on any single source and limit the impact of an outage.

Configurations and rates

Current configurations, rates, and locations are listed on the pricing page. Every Kinesis-supplied configuration is quoted all-in, Grid included.

Need a particular configuration or a larger deployment? Tell us the hardware, quantity, location, and timing you need.

Included with every deployment

Your compute comes ready for the Grid.

Every Kinesis-supplied deployment includes a private Grid: one control plane that makes decisions day and night across machines with different costs, speeds, failure habits, and owners.

Deploy applications

Five ways in, one runtime, one placement engine, one set of operations, one bill. Connect a repository, point us at your registry, hand us a Dockerfile, push an image, or start from a template.

Getting applications on → (planned)

Use capacity well

Allocate to approved users. Split a card into slices so light users share it without getting in each other's way. Queue by fair share when the machines fill up. Warn a quiet session, then take the machine back.

Reclaiming idle capacity →

Keep work running

Kinesis handles placement, scaling, load balancing, failover, migration, and configuration consistency. When a machine fails, the system spots it, walls it off, and moves the work to healthy machines without breaking your rules about where work may run.

When machines fail → (planned)

See every decision

Ask where a job ran last Tuesday and why the system chose that machine, and get an answer.

The decision record → (planned)

Fractional dedicated compute

Give each workload the GPU capacity it needs.

Smaller workloads get an isolated slice of a real GPU, sized to the work, on dedicated hardware. Approved users share the supplied capacity; your administrator controls allocation and access. Not a container dropped onto a machine shared with strangers.

Fractional dedicated compute →

Yours and ours

Start on ours. Bring yours when you are ready.

Because the boundary belongs to you, our machines and yours go together in any amount, in any order, and you can change your mind later without moving to a different platform. A team can start on our machines, then bring its own hardware into the same Grid, and its applications never notice.

SAME PRODUCT

Find the capacity for your next workload.

Start with your workload requirements. Kinesis brings the compute and the platform to run it. Every Kinesis-supplied configuration is quoted as one all-in hourly or monthly price, with your private Grid included.