Limited access
Elastic compute for AI developers
Elastic compute for AI developers
Elastic compute for AI developers
Access NVIDIA GPUs in minutes for training, fine-tuning, and inference.
No contracts. No flaky machines.
Access NVIDIA GPUs in minutes for training, fine-tuning, and inference.
No contracts. No flaky machines.
Trusted by
Reserved instances
Reserve compute when you need it.
Sell it back when you don't.
Guaranteed performance and reliability for training, fine-tuning, and inference
Reserved instances
Reserve compute when you need it.
Sell it back when you don't.
Guaranteed performance and reliability for training, fine-tuning, and inference
Reserve a few hours or a few weeks
Provision NVIDIA GPU clusters with 3200Gbps InfiniBand for as little as 3 hours. Guarantee the capacity you need for critical workloads.
Resell capacity you aren't using
Reserved more than you need? Put unused instances in the spot pool to generate credits and take them back whenever you need them.
Reserve a few hours or a few weeks
Provision NVIDIA GPU clusters with 3200Gbps InfiniBand for as little as 3 hours. Guarantee the capacity you need for critical workloads.
Resell capacity you aren't using
Reserved more than you need? Put unused instances in the spot pool to generate credits and take them back whenever you need them.
Spot instances
Burst compute when the price is right.
20x your price-performance.
Cost-efficient compute for flexible training, fine-tuning, and inference
Spot instances
Burst compute when the price is right.
20x your price-performance.
Cost-efficient compute for flexible training, fine-tuning, and inference
Spot instances
Burst compute when the price is right.
20x your price-performance.
Cost-efficient compute for flexible training, fine-tuning, and inference
Set a max price that suits your task
Set a max price that suits your task
Set a maximum compute price that makes sense for your workload and Foundry gives you compute whenever the market price is lower. Place orders at a higher price to scale your compute quickly or set lower prices to save on workloads that can be delayed.
Set a maximum compute price that makes sense for your workload and Foundry gives you compute whenever the market price is lower. Place orders at a higher price to scale your compute quickly or set lower prices to save on workloads that can be delayed.
Programmatically scale up & down
Programmatically scale up & down
Place spot orders via API to automate scaling and provision instances to a Kubernetes cluster to simplify orchestration. Gracefully handle startup and preemption with custom scripts, disk state saving, auto-mounting persistent storage, and more.
Place spot orders via API to automate scaling and provision instances to a Kubernetes cluster to simplify orchestration. Gracefully handle startup and preemption with custom scripts, disk state saving, auto-mounting persistent storage, and more.
From our partners and customers
From our partners and customers
“When we believe we could benefit from additional compute, we just turn it on. When we need to pause to study our results and design the next experiment, we turn it off.
Because we aren’t locked into a long-term contract, we have the flexibility to experiment with a variety of GPUs and empirically determine how to get the best price-performance for our workload.”
Matt Wheeler, AI Research Engineer
“Foundry Cloud Platform has accelerated science at Arc. Our machine learning work brings demanding performance infrastructure needs, and Foundry delivers.
With Foundry, we can guarantee that our researchers have exactly the compute they need, when they need it, without procurement friction.”
Patrick Hsu, Co-Founder and Core Investigator
“There’s just nothing out there comparable to the prices we got per GPU hour than we did on Foundry. Even though I was on a tight deadline I never had to worry about compute because I knew the machines I wanted would be available on Foundry’s cloud platform.”
Keyon Vafa, Machine Learning Researcher
“When we believe we could benefit from additional compute, we just turn it on. When we need to pause to study our results and design the next experiment, we turn it off.
Because we aren’t locked into a long-term contract, we have the flexibility to experiment with a variety of GPUs and empirically determine how to get the best price-performance for our workload.”
Matt Wheeler, AI Research Engineer
“Foundry Cloud Platform has accelerated science at Arc. Our machine learning work brings demanding performance infrastructure needs, and Foundry delivers.
With Foundry, we can guarantee that our researchers have exactly the compute they need, when they need it, without procurement friction.”
Patrick Hsu, Co-Founder and Core Investigator
“There’s just nothing out there comparable to the prices we got per GPU hour than we did on Foundry. Even though I was on a tight deadline I never had to worry about compute because I knew the machines I wanted would be available on Foundry’s cloud platform.”
Keyon Vafa, Machine Learning Researcher
Built for AI engineers, researchers, and scientists
Built for AI engineers, researchers, and scientists
Run custom scripts on startup and via SSH
Manage instances from your CLI or via API
Access co-located storage (no ingress/egress fees)
Run custom scripts on startup and via SSH
Manage instances from your CLI or via API
Access co-located storage (no ingress/egress fees)
NVIDIA Tensor Core GPUs
Access NVIDIA H100s, A100s, A40s, and A5000s without a contract
1.6 Tbps Infiniband
Optimize distributed training with high-performance networking
Native Kubernetes
Simplify workload orchestration and horizontal scaling with Kubernetes
NVIDIA Tensor Core GPUs
Access NVIDIA H100s, A100s, A40s, and A5000s without a contract
1.6 Tbps Infiniband
Optimize distributed training with high-performance networking
Native Kubernetes
Simplify workload orchestration and horizontal scaling with Kubernetes
NVIDIA Tensor Core GPUs
Access NVIDIA H100s, A100s, A40s, and A5000s without a contract
1.6 Tbps Infiniband
Optimize distributed training with high-performance networking
Native Kubernetes
Simplify workload orchestration and horizontal scaling with Kubernetes
Enterprise-ready scale and security
Enterprise-ready scale and security
SOC2 Type II certified
Built to the highest standards of security from the ground up.
Multi-layer security
All of our servers are located in Tier 3 and 4 data centers.
Redundant & fault tolerant
We handle burn-in, observability, and alerting so you don’t have to.
Granular access control
Set user roles and permissions for managing compute.
High-performance storage
We offer block storage and file shares with no ingress or egress charges.
Professional services
We are a team of AI researchers and infrastructure experts.
SOC2 Type II certified
Built to the highest standards of security from the ground up.
Multi-layer security
All of our servers are located in Tier 3 and 4 data centers.
Redundant & fault tolerant
We handle burn-in, observability, and alerting so you don’t have to.
Granular access control
Set user roles and permissions for managing compute.
High-performance storage
We offer block storage and file shares with no ingress or egress charges.
Professional services
We are a team of AI researchers and infrastructure experts.