Ollama的本地AI赌注：GitHub 17.6万星标，6500万美元B轮，890万开发者

做的是什么生意

An open-source tool that lets developers find, download and run open-weight AI models on their own hardware with one command — 'Docker for AI' — plus a paid cloud for models too big for local machines, billed by GPU time.

启动资金：$15M Series A led by Benchmark (Peter Fenton joined the board); $65M Series B led by Theory Ventures with Benchmark, 8VC, Y Combinator, Pace Capital, 49 Palms, GTMFund and angels; $88M total

起因

Jeffrey Morgan and Michael Chiang had built Kitematic, which Docker acquired, and worked on Docker Desktop — the playbook for making developer tooling ubiquitous. When open-weight models arrived in 2023 they saw researchers could use them but programmers couldn't: installs were painful and fragmented. They launched Ollama in 2023 to make running a model locally feel like pulling a container, and it spread rapidly among developers.

经过

Ollama became the de facto local model runtime with 67,000 community integrations, day-zero access to new models from Meta, Google DeepMind, Mistral and MiniMax, and hardware partnerships with NVIDIA, Intel, AMD and Qualcomm. In 2025 its paid cloud drew an 'enshittification' backlash from free-tier fans; Morgan argued the cloud funded the mission. Agentic coding tools like OpenClaw drove a usage surge in January 2026, and usage doubled by July.

结果

Still running and scaling: $65M Series B announced July 9, 2026, led by Theory Ventures; 8.9M monthly developers, 85% Fortune 500 penetration, 14 employees; free local product intact plus paid cloud tiers up to $100/month billed by GPU time.

可借鉴之处

Ollama靠分发而非模型取胜：每个模型实验室和硬件供应商都需要的一键安装，之后通过云算力变现——这个平台太方便，无法被替代。