SaladCloud
Distributed GPU cloud using idle consumer hardware
SaladCloud aggregates idle GPUs from hundreds of thousands of individual users and enterprises to provide low-cost compute. Founded in 2018, it offers container orchestration for AI inference and workloads across a global distributed network.
- Over 450,000 GPUs have contributed to the network since 2018
- Nodes are distributed across more than 190 countries worldwide
- Pricing starts at $0.015 per GPU-hour for on-demand workloads
- Uses hypervisor isolation to secure containers on consumer hardware
- Joined the Render Network as a subnet in 2026
What it is
SaladCloud is a distributed computing platform that aggregates idle graphics processing units from consumer PCs and enterprise workstations. The company operates a two-sided network where individuals, known as Salad Chefs, share their hardware capacity in exchange for rewards. Businesses deploy containerized workloads to this global pool via the Salad Container Engine. This model allows for significantly lower costs compared to traditional data centres by utilizing existing hardware rather than building new infrastructure. The service supports AI inference, image generation, and other parallelizable tasks across more than 190 countries.
How it came about
The company incorporated in 2018 with seed investment from Kickstart Seed Fund and Royal Street Ventures. It began as a crypto experiment before pivoting to general compute orchestration. By 2019, the network claimed supercomputer status, surpassing Cray's Jaguar in performance metrics. A seed-plus round of $3.2 million followed in 2020. In 2022, Salad raised $17 million in Series A funding co-led by Left Lane Capital and Origin Ventures. That same year, it launched the Salad Container Engine. Recent developments include joining the Render Network as a subnet in 2026 and introducing reserved hardware options.
The compute
The platform relies on the Salad Container Engine to schedule and monitor Docker workloads across heterogeneous infrastructure. It uses hypervisor isolation to ensure security, running containers in partitioned Linux virtual machines. The network primarily features NVIDIA RTX-class consumer GPUs, ranging from the RTX 3060 to the RTX 5090, along with selected AMD cards. A proprietary trust system scores node reliability. For enterprise needs, Salad Dedicated offers reserved bare metal and virtual machines, including upcoming NVIDIA Blackwell RTX PRO 6000 units. The system handles automatic reallocation if a node becomes unavailable.
Where the GPUs are
Compute resources are distributed globally, with nodes located in over 190 countries. This decentralized approach means there is no single central data centre footprint for the community cloud. Instead, the infrastructure exists wherever participants have idle hardware. The company also offers Private Cloud solutions, allowing organizations to run the orchestration stack on their own internal devices. This enables companies to pool idle workstation GPUs within their own facilities while keeping data local. Reserved hardware through Salad Dedicated provides more traditional, contracted capacity.
Customers and deals
Production customers include Civitai, which uses over 600 GPUs to serve millions of images daily. Blend reported running three times the scale at half the cost of previous providers. Other users include Klyne.ai, Undetectable AI, and Pareto. These clients leverage the platform for inference tasks such as image generation, voice transcription, and object detection. The company states that its pricing can be up to 90 per cent lower than hyperscale cloud providers for suitable workloads. Engineers work directly with these teams to optimize container tuning and architecture.
Money and ownership
Salad has raised venture capital from firms including Left Lane Capital, Origin Ventures, Kickstart Seed Fund, and Royal Street Ventures. The Series A round in 2022 amounted to $17 million. An earlier seed-plus round in 2020 raised $3.2 million. The business model relies on paying users for their idle compute power while charging businesses for usage. Pricing starts at $0.015 per GPU-hour for community cloud resources. The company claims this structure creates a lower cost base by avoiding the capital expenditure required for new data centre construction.
Its place in AI
Salad positions itself as a solution to the compute crisis, arguing that centralized providers ration supply and set high prices. By activating consumer hardware, it aims to democratize access to AI computing power. The network joined the Render Network as a subnet in 2026, expanding its reach within the decentralized compute ecosystem. It serves as an alternative for workloads that do not require sub-10-millisecond latency or NVLink training. The platform supports open-weight models via an OpenAI-compatible API, allowing developers to integrate easily without changing their tooling.
Questions and criticism
The model relies on interruptible supply, meaning nodes may disappear unexpectedly, though the engine handles reallocation. Workloads requiring strict data residency guarantees or multi-GPU training links are not suitable for the community cloud. Security is managed through hypervisor isolation and host-intrusion detection, but the distributed nature involves trusting remote hardware. The company notes that its SOC 2 Type I compliance addresses some enterprise security requirements. Users must accept that their containers run on personal devices, which may vary in stability compared to dedicated server racks.
Questions people ask about SaladCloud
What types of GPUs are available on SaladCloud?
The network primarily uses NVIDIA RTX-class consumer GPUs, from the RTX 3060 to the RTX 5090. Selected AMD cards are also included. Enterprise customers can access reserved data-center GPUs like the RTX PRO 6000 Blackwell through Salad Dedicated contracts.
How does SaladCloud ensure security on consumer devices?
Containers run inside Linux virtual machines isolated by a hypervisor. All images and traffic are encrypted. If a host attempts to access the container, intrusion detection destroys the environment and blacklists the machine. The platform is SOC 2 Type I compliant.
Is SaladCloud suitable for AI model training?
It is optimized for inference and parallelizable workloads. It is not suitable for tasks requiring sub-10-millisecond latency, multi-GPU NVLink training, or strict data residency. For these needs, Salad offers reserved hardware or private cloud options.
How do individuals earn money by sharing their GPUs?
Users, called Salad Chefs, install a desktop app that runs when their PC is idle. They earn Salad Balance based on the compute provided. This balance can be redeemed for PayPal cash, gift cards, games, and other rewards.
Good reading and sources
There is no English Wikipedia article on SaladCloud yet, so the facts on this page come from the sources listed above: salad.com/about, salad.com/. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryNeocloud wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.