SaladCloud

SaladCloud (operated by Salad Technologies) runs a distributed GPU cloud built on consumer-grade hardware rather than datacenter-class accelerators. Home PC owners called "Salad Chefs" install a desktop client that rents out idle GPU cycles in exchange for gift cards, game credits and streaming subscriptions; SaladCloud resells that aggregated capacity to AI companies as a low-cost inference and batch-compute layer. The company is headquartered in Salt Lake City and was founded in 2018, initially as a way for gamers to monetize idle machines through cryptocurrency mining before pivoting to AI/ML workloads.

The positioning is deliberately opposite to the H100/B200 hyperscaler-adjacent neoclouds like CoreWeave, Crusoe or Lambda. SaladCloud does not own datacenters, does not report a megawatt footprint, and does not carry Nvidia Hopper or Blackwell inventory in any meaningful quantity; its fleet is dominated by consumer RTX cards (3060 through 4090-class) contributed by tens of thousands of individual "Chef" nodes. That gives it a structural cost advantage for embarrassingly parallel inference, transcription, image generation and rendering jobs where a single H100 is overkill, at the price of no SLA-grade uptime on any individual node and no support for large-model training that needs high-bandwidth GPU-to-GPU interconnect.

Public disclosures on the company are thin: SaladCloud has not announced a priced funding round since its 2022 Series A, has not disclosed total GPU count or revenue, and remains a small private company (LinkedIn shows roughly 40-50 employees as of 2026). Its competitive category inside the "neocloud" pack is the long-tail / crowd-sourced compute layer alongside Vast.ai and the compute side of decentralized networks like Akash and io.net, rather than the hyperscale training clouds.