Show HN: Salad, a distributed cloud for AI (like Airbnb for GPUs)
The orchestration cloud
Salad orchestrates the world's idle GPUs.
One orchestration layer for containers, tokens, and GPUs, running across 16,000+ daily active GPUs in 190+ countries, or privately inside your own organization. Up to 90% less than hyperscaler pricing.
No contracts, no minimums, per-second billing on SaladCloud. Reserved capacity and private deployments available for enterprise.
# Deploy a container group to 40 GPUs across the network curl -X POST https://api.salad.com/api/public/organizations/acme/projects/prod/containers \ -H "Salad-Api-Key: $SALAD_API_KEY" \ -H "Accept: application/json" \ -H "Content-Type: application/json" -d '{ "name": "inference", "display_name": "Inference", "container": { "image": "ghcr.io/acme/inference:latest", "resources": { "cpu": 4, "memory": 8192, "gpu_classes": ["9998fe42-04a5-4807-b3a5-849943f16c38"] }, "priority": "high" }, "replicas": 40, "autostart_policy": true, "restart_policy": "always" }' # → 40 replicas scheduled onto RTX 4090 nodes worldwide. Billing starts only when each instance is running.
Running in production at
CivitaiBlendKlyne.ai Undetectable AIParetoDatasembly
16,000+daily active GPUs
190+countries with nodes
90%lower cost vs. hyperscalers
$0.015per GPU-hour, starting price
Platform and deployment options
One orchestration layer, across every pool.
Use a managed token API, deploy containers through Salad Container Engine, or choose the infrastructure that fits. Community Cloud is the on-demand GPU pool; Dedicated provides reserved hardware; Private Cloud runs the same orchestration inside your organization.
Managed token API
AI Gateway
An OpenAI-compatible endpoint for open-weight models, served from the network. Pay per token or request a flat monthly plan; your data is never used for training.
openai apiqwen262k ctx
Core orchestration
Salad Container Engine
The software layer that schedules, monitors, and reallocates Docker workloads across distributed GPU pools. Use it on Community Cloud, Dedicated, or your own fleet.
docker10,000+ replicasfailover
Dedicated infrastructure
Salad Dedicated
Reserved bare metal and virtual machines on contract. RTX PRO 6000 Blackwell first, with thousands of cards coming online.
rtx pro 6000bare metalcontracts
On-demand infrastructure
Community Cloud
The distributed pool of idle consumer GPUs that Salad Container Engine deploys onto, with priority pricing and capacity across 190+ countries.
rtx 5090rtx 4090from $0.015/hr
Why Salad
Two competencies, nine years, one platform.
Most clouds operate uniform racks in a few data centers. Salad was built to turn diverse hardware across thousands of locations into dependable, usable infrastructure.
Orchestration across distributed infrastructure
Salad Container Engine schedules OCI containers onto nodes it has never seen before, verifies them, isolates them behind a hypervisor, streams logs, and reallocates work the moment a node disappears. It runs identically on consumer PCs, workstation fleets, and data-center servers.
Orchestration across idle GPUs
450,000+ GPUs have contributed to the network since 2018. A proprietary trust rating scores node performance and availability; priority pricing matches interruptible supply to workloads designed for it. The result is a GPU pool with exceptional scale, utilization, and economics.
Economics no data center can match
The hardware already exists. Salad coordinates supply instead of financing new data-center capacity, creating a structurally lower cost base and helping customers run substantially more work within the same budget.
Hands-on engineering support
Production customers work directly with Salad engineers on architecture, container tuning, and cost. Volume deployments and committed contracts receive tailored support and pricing.
Where it runs
SaladCloud, or a private cloud on your own devices.
SaladCloud
Deploy on the global network. Community Cloud for the lowest cost anywhere; Salad Dedicated for reserved capacity on contract. Same API, same portal, same orchestration.
Private Cloud
Run Salad's orchestration stack inside your organization. Pool the idle GPUs in workstations and servers you already own into a private, OpenAI-compatible token service. Company data never leaves company devices.
Optimized per workload
Every modality, tuned for the network.
Published benchmarks, not promises. Each figure links to the methodology.
Image generation
5,243 Flux.1-Schnell images per dollar. Civitai serves 10 million images a day on 600+ Salad GPUs.
Transcription
95.1% accuracy on CommonVoice, the top of the benchmark, from $0.16 per audio hour.
LLM inference
Flat hourly GPUs for agents and RAG, or per-token pricing through AI Gateway from $0.06 per million input tokens.
Computer vision
309,000 images tagged per dollar with RAM++; YOLOv8 object detection for 73% less than Azure.
Batch & simulation
1,000,000 hours of audio transcribed for $1,260; GROMACS at 90%+ below data-center pricing.
Customer stories
Teams ship on Salad.
Civitai
"By switching to SaladCloud, we are serving inference on over 600 consumer GPUs to deliver 10 Million images per day and training more than 15,000 LoRAs per month. SaladCloud not only had the lowest GPU prices in the market but also offered us incredible scalability."
Justin Maier · Founder & CEO, CivitaiRead the case study →
Blend
"On SaladCloud's consumer GPUs, we are running 3X more scale at half the cost of A100s on our local provider and almost 85% less cost than the two major hyperscalers we were using before. I'm not losing sleep over scaling issues anymore."
Jamsheed Kamardeen · CTO, BlendRead the case study →
Klyne.ai
"If you want access to 1000s of GPUs, you can get them on SaladCloud for better cost-efficiency. Salad is also really customer friendly, something you cannot get from a larger cloud provider as a startup."
Zachary Lawrence · CEO, Klyne.aiRead the case study →
From the blog
How to build on Salad.
Benchmarks, tutorials, and product updates.
What kind of GPUs are on SaladCloud?
NVIDIA RTX-class consumer GPUs, from the RTX 3060 through the RTX 5090 (32 GB), plus selected AMD classes. Every GPU is benchmarked before it is admitted and continuously scored by our trust rating. Data-center and workstation GPUs, starting with the RTX PRO 6000 Blackwell, are available on contract through Salad Dedicated.
How is this different from a spot instance?
Community Cloud nodes are interruptible, like spot, and priced by priority tier (High, Medium, Low, Lowest). Unlike spot, Salad Container Engine handles the consequences: it reallocates your workload to another node of the same class automatically, bills only for running time, and never charges for cold start or image download.
How does security work on a network of other people's computers?
Your container runs inside a Linux virtual machine partitioned from the host by a hypervisor, with image, configuration, and traffic encrypted in transit and at rest. Host-intrusion detection destroys the environment and blacklists the machine if anything attempts to access it. Salad is SOC 2 Type I compliant. Read the security overview.
What workloads are not a fit?
Anything that needs sub-10-millisecond latency, multi-GPU NVLink training, or data-residency guarantees that a consumer network cannot provide. For those requirements, Salad Dedicated offers reserved hardware and Private Cloud runs on your own infrastructure. Our engineers are available to support the deployment choice and architecture.
Why do people share their GPUs with Salad?
They earn. Salad Chefs run our desktop app when their PC is idle and redeem Salad Balance for PayPal, games, gift cards, and more. Since 2018, more than 450,000 GPUs have contributed compute to the network. Learn about becoming a Chef.
Can we run Salad's technology inside our own company?
Yes. Private Cloud deploys Salad's orchestration stack on your infrastructure and pools the idle GPUs in your existing workstations and servers into a private, OpenAI-compatible endpoint. Prompts, completions, and embeddings never leave your network.
Explore Salad today.
Deploy on SaladCloud, or talk to sales about reserved and private options.