Spaces
Hugging FaceSpaces hosts interactive machine-learning apps on Hugging Face infrastructure, built with the Gradio or Streamlit SDKs, an arbitrary Dockerfile, or static HTML and JavaScript. Hardware runs from a free CPU tier through a shared ZeroGPU pool to hourly dedicated GPUs, alongside persistent storage and an SSH dev mode. Hub integration deploys a Space from a model card, and a Space can be exposed as an MCP server or API endpoint.
The declared GitHub artifact was gradio-app/gradio, removed 2026-08-14: it is an app SDK that runs anywhere, is already gradio's own artifact, and would have banded Spaces on Gradio's stars. Spaces publishes no countable channel of its own. Verified 2026-08-13 via the Spaces documentation and pricing page.
Openness
1 medium confidence- license
- Proprietary
- source
- closed
- platform
- proprietary(HF-managed hosting/orchestration, no self-hostable Spaces)
- open-part
- app SDKs only (Gradio Apache-2.0, Streamlit) which are separate, portable OSS products
- hardware-tiers
- CPU free + paid GPU upgrades
Closed managed hosting service. Apps are built with OSS SDKs (Gradio etc.) that run anywhere, but Spaces itself - the managed host/orchestration - is proprietary and not self-hostable; those SDKs are separate products, not a Spaces open core. Reclassified from open_core (borderline). The Hub docs describe Spaces only as apps hosted on Hugging Face infrastructure - Gradio, arbitrary Dockerfile, static HTML/JS, with GPU hardware upgrades, dev mode and persistent storage - and document no way to run the Spaces host anywhere else. The pricing page sells that host as metered HF compute (CPU Basic free, CPU Upgrade $0.03/hr, ZeroGPU free with a PRO quota, then hourly T4/L4/A100 tiers), and the terms of service reserve ownership of the service outright - "we retain ownership of all of our intellectual property rights related to the Website and the Services" - which is what the Proprietary license part records. No source repository or self-host distribution for Spaces itself exists.
- https://huggingface.co/docs/hub/spaces recorded 2026-08-13
Spaces hosts apps via Gradio (OSS SDK), Streamlit, arbitrary Dockerfile, or static HTML on HF-managed infra with paid GPU upgrades
- https://huggingface.co/pricing recorded 2026-08-13
Spaces Hardware table - the Spaces host is sold as metered Hugging Face compute. CPU Basic (2 vCPU / 16 GB) free, CPU Upgrade $0.03/hr, ZeroGPU free on a shared Blackwell pool with an 8x quota for PRO, then hourly Nvidia T4 / L4 / A100 tiers. PRO and Team plans are described as hosting "your own ZeroGPU, Gradio & Docker Spaces on compute" - HF compute, with no self-hosted option offered anywhere on the page.
- https://huggingface.co/terms-of-service recorded 2026-08-13
Intellectual Property section - "Proprietary Rights. We retain ownership of all of our intellectual property rights related to the Website and the Services, including all improvements to the Services. All materials that we produce, including the Website, design, code, graphics, interfaces, trademarks" are Hugging Face's. Spaces is one of the Services, so the platform itself is governed by these proprietary terms rather than by an open license.
Adoption
5 medium confidenceHugging Face’s homepage states that Spaces hosts more than 1M AI applications, on a platform used by more than 50,000 organizations. 1M+ hosted apps plus the broad end-user traffic to those demos puts served reach into the >10M band. The headline figure is the vendor’s own and per-Space user totals are not individually disclosed, so confidence is medium. There is no countable artifact behind the level, and none is claimed: a Space is not a package, and the 1M+ figure is Hugging Face’s own count of hosted apps rather than a measurable distribution channel, so the band cannot be recomputed from a feed. The source below is digested so the claim can at least be checked again and will age; a route to a countable Spaces signal remains an open gap.
- https://huggingface.co recorded 2026-08-13
Spaces panel closes with "Browse 1M+ applications"; the customer band reads "More than 50,000 organizations are using Hugging Face" over Ai2, Meta, Amazon, Google, Intel, Microsoft. No per-Space or aggregate visitor figure is published.
Capability
3 medium confidenceExcellent for hosting/demoing ML apps with broad SDK and GPU support, but it is an app-hosting platform rather than a hardened arbitrary-untrusted-code sandbox (container, not microVM isolation), and free Spaces sleep. Mid-tier within the deployment sandbox lens. Nothing in the Hub docs or the pricing page claims microVM or hardened multi-tenant isolation, and the hardware ladder still starts at a free 2 vCPU CPU Basic tier. Spaces as MCP servers, agent tools and API endpoints now have their own doc pages, which widens the surface without moving the isolation ceiling the band rests on.
- https://huggingface.co/docs/hub/spaces recorded 2026-08-13
Gradio/Streamlit/Docker/static hosting, GPU hardware upgrades, persistent storage and dev mode
- https://huggingface.co/pricing recorded 2026-08-13
Spaces Hardware ladder - free CPU Basic, $0.03/hr CPU Upgrade, free ZeroGPU shared pool, then hourly T4 / L4 / A100 tiers
Verified 2026-08-13