Two competencies. Three ways to deploy. One vendor.
Salad knows how to orchestrate workloads across distributed infrastructure, and how to turn idle GPUs into a cloud. Enterprises consume that in whichever shape fits: the public network, reserved hardware on contract, or a private deployment on their own fleet.
Match the pool to the workload.
Most enterprise customers end up with more than one. The API, portal, and support team are the same across all three.
| Community Cloud | Salad Dedicated | Private Cloud | |
|---|---|---|---|
| Where it runs | Salad's global network of consumer GPUs | Reserved, single-tenant servers across Salad's distributed footprint | Your own workstations and servers |
| GPUs | RTX 3060–5090, 8–32 GB | RTX PRO 6000 Blackwell, 96 GB | Whatever your fleet already has |
| Interruption | Interruptible, auto-reallocated | None | Yields to the primary user |
| Data boundary | Encrypted, hypervisor-isolated on third-party nodes | Single-tenant, reserved hardware | Never leaves your network |
| Pricing | Per second, published rates, volume discounts | Contract, by quote | License per deployment |
| Best for | Inference and batch at the lowest cost | Sustained serving, large models, training | Internal LLM access with data sovereignty |
| Learn more → | Learn more → | Learn more → |
Designed for hostile hosts. Comfortable in friendly ones.
Salad's security model was built for the hardest case, running customer workloads on machines Salad does not own. The same controls carry over to Dedicated and Private Cloud.
SOC 2 Type I
Audited against AICPA standards. Reports and policies available through the Trust Center.
Hypervisor isolation
Workloads run in a Linux VM partitioned from the host by Hyper-V, inside an OCI container. Isolation protects both directions.
Encryption everywhere
Images, configuration, environment variables, queue payloads, and gateway traffic encrypted in transit and at rest with modern cipher suites.
Host-intrusion detection
Any attempt by a host to access a workload's environment destroys the environment and blacklists the machine. Runtime detection is being extended with Falco.
Trust rating
Every node earns a rating from behavior and performance history. Only nodes in good standing receive workloads.
Region constraints
Pin workloads to countries or regions when regulation requires it, or take the question off the table entirely with Private Cloud.
Plain statement of fit: workloads with strict regulatory requirements or a high data-to-compute ratio are often not suited to Community Cloud. We will tell you that up front and point you to Dedicated or Private Cloud instead.
Engineers, not ticket queues.
Named engineering contact
Production customers get a Salad engineer who knows their workload, plus a shared channel for day-to-day questions.
Architecture and migration
Container tuning, reallocation strategy, queue design, and cost modeling before you deploy, and migration guides from the APIs you're leaving.
Commercial flexibility
Volume discounts above 100 GPUs on Community Cloud, committed-spend agreements, contract terms on Dedicated, monthly invoicing for enterprise accounts.
Documentation and status
Public docs, API reference, SDKs, and a live status page. No surprises is a brand value, not a slogan.
What customers report.
"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, Blend"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."
Justin Maier · Founder & CEO, CivitaiUndetectable.ai reports saving over $50,000 a month on inference after moving to SaladCloud.
Case studyRead on the blog →Bring us the workload. We'll bring the right pool.
A 30-minute call with an engineer is usually enough to size the deployment and quote it.