Free open-source self-hosted runtime (the only real free offer; available to students and startups alike) high confidence
GPUStack core is free, open-source, Apache 2.0 licensed (NOT MIT as the task's 'known so far' said — both gpustack.ai and the GitHub repo state Apache 2.0 today). One-line Docker install (server: sudo docker run -d --name gpustack -p 80:80 --volume gpustack-data:/var/lib/gpustack gpustack/gpustack; worker: separate privileged Docker command with --server-url and a generated --token). Includes Ollama-style local web UI + OpenAI/Anthropic-compatible API, model deployment, and supports heterogeneous accelerators (NVIDIA; AMD/ROCm, Ascend, Hygon, MetaX, Moore Threads, Cambricon, Iluvatar, T-Head PPU per the site). Free forever, no expiry. No hosted free tier, no trial credits, no student/education offer, no startup program exists on the official site (verified /, /enterprise, /blog, GitHub, docs quickstart on 2026-10-22; /cloud/ and /pricing/ are 404). Sources read today: https://gpustack.ai, https://gpustack.ai/enterprise, https://gpustack.ai/blog/, https://gpustack.ai/blog/introducing-gpustack/, https://github.com/gpustack/gpustack, https://docs.gpustack.ai/latest/quickstart/
- Eligibility
- Anyone — students, startups, enterprises, individuals. No .edu email, no account, no company age or funding-cap requirements, no region limits. Credit card NOT required; no commercial license, no login/account to set up (local 'admin' user with a system-generated password).
- How to claim
- Nothing to claim or sign up for: clone/download the open-source code (https://github.com/gpustack/gpustack) and run the Docker one-liners from https://docs.gpustack.ai/latest/quickstart/. Worker nodes need Linux + at least one GPU (docs quickstart covers NVIDIA; site advertises AMD/ROCm and Chinese accelerators too; macOS/Windows can host the server via Docker Desktop, workers are Linux-only; Windows via WSL2).
- Expiry
- Indefinite — Apache 2.0 license is perpetual; no sunsetting date for the open-source product.
- Gotchas
- The 'free compute' is BYO-hardware: you pay for your own GPU, electricity, and (for model pulls from hub sources) data egress; there is no GPUStack-hosted free GPU pool. Workers are Linux-only — no macOS worker nodes. RBAC, SSO (OIDC/SAML/LDAP), token quotas, metering & billing, white-label, HA and multi-tenancy are Enterprise features sold via 'Talk to Sales'/'Book a Demo' (https://gpustack.ai/book-demo) with no published pricing — do not assume free. The June 2026 'Introducing GPUStack' blog repositions the product as an 'enterprise AI infrastructure platform' with 'MaaS and GPUaaS under one control plane' (https://gpustack.ai/blog/introducing-gpustack/), so a commercial hosted GPUaaS is signaled, but no public self-serve page, pricing, or free tier for it was found (gpustack.ai/cloud and gpustack.ai/pricing return 404). WebSearch was unavailable this session (session budget exhausted), so third-party sources (funding news, off-site discount resellers) could not be checked.
- Source
- gpustack.ai