Salad

Salad operates SaladCloud, a distributed cloud computing platform that aggregates idle consumer GPUs contributed by individual PC owners and resells that capacity to businesses running GPU-heavy workloads. The company markets the network as a low-cost alternative to hyperscale cloud providers, advertising more than 60,000 daily active GPUs with pricing starting around $0.02 per hour, and describes a footprint of over one million individual nodes across more than 180 countries.

The platform is container-based: customers package applications for the Salad Container Engine (SCE) and Salad handles scheduling, scaling and node management rather than exposing virtual machines or individual instances. Targeted workloads include text-to-image generation, text-to-speech, speech-to-text transcription, computer vision and large language model inference, alongside batch data processing, HPC and rendering queues. Salad also offers a packaged Salad Transcription API built on Whisper Large v3, with LLM-based insights using Llama 3 8B. Supply is sourced through a consumer-facing side of the business in which GPU owners install software to share spare capacity in exchange for rewards.

Business model

Two-sided marketplace: individual GPU owners share spare compute capacity in return for rewards, while business customers pay usage-based fees to run containerized workloads on that pooled capacity. Salad presents itself as a fully managed container service, handling orchestration across the node network.

Usage-based pricing for GPU compute and managed inference services, with no pre-paid contracts or commitments and discounts offered on recurring plans. Examples of published pricing include GPU capacity from $0.02/hour, LLM deployment from $0.04/hour, 7-billion-parameter model serving at $0.22/hour, and transcription from $0.10 per hour of audio.

Traction

Company-stated figures include 60,000+ daily active GPUs, over one million individual nodes, and hundreds of customers. Customer-reported results include Civitai serving inference on over 600 consumer GPUs to deliver 10 million images per day and training more than 15,000 LoRAs per month, and Blend running 3x more scale at half the cost of A100s from its prior provider.

Latest developments

Recent published work covers benchmarks and cost analyses for Qwen 3.5 small models, use of the Cline coding agent on SaladCloud, an automated news bot built with OpenClaw, Flux.1-Schnell image generation on RTX 5090 hardware optimized with TheStage AI's ANNA, and an integration with the Render Network.