San Francisco Compute operates a marketplace for on‑demand GPU‑accelerated virtual machines, letting users provision anywhere from a single node to hundreds of GPUs for any length of time. Pricing is transparent, pay‑as‑you‑go (average $1.98 per H100 GPU‑hour) with options for reserved or interruptible usage, and nodes are managed via a simple CLI or API that supports custom UEFI images and cloud‑init scripts.
Funding
$12M raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
2OACFounders
Product
Problem
Accessing large-scale GPU compute typically requires long-term contracts, upfront hardware investment, or navigating complex cloud provider pricing, which limits flexibility for researchers and companies with variable workloads.
Solution
San Francisco Compute operates a marketplace that lets users purchase GPU‑accelerated virtual machines on demand, scaling from a single node to hundreds of GPUs for any duration. Users interact via a simple command‑line interface or API to reserve nodes at a market price, with options for guaranteed (reserved) or interruptible (auto‑reserved) usage. Pricing is transparent and billed per GPU‑hour, and nodes can be released or sold back at any time, eliminating lock‑in. The platform provides UEFI‑bootable VMs, supports custom OS images and cloud‑init scripts, and abstracts data transfer costs by keeping workloads within the provider’s vetted clusters.
Target Audience
Primary customers are machine‑learning researchers, data‑science teams, and AI‑focused startups that require elastic, high‑performance GPU compute for training, inference, or experimentation.
Features
- Pay‑as‑you‑go pricing model with average $1.98 per H100 GPU‑hour and market‑based bidding for auto‑reserved nodes
- CLI (`sf`) and REST API for rapid provisioning, scheduling, and management of GPU VM nodes
- Reserved nodes guarantee access for a specified time window; auto‑reserved nodes enable cost‑effective, interruptible workloads
- Support for custom UEFI bootable images and cloud‑init scripts to configure environments at launch
- Zone selection for physical colocation of nodes, with future InfiniBand support for high‑speed interconnects
- Transparent SLA backed by vetted data‑center providers, with hot‑swap replacements or refunds for hardware failures