The open-source AI tool that lets developers run powerful models on their own machines just closed a $65 million Series B round led by Theory Ventures. Founder and CEO Jeff Morgan confirmed the raise, which follows a $15 million Series A from Benchmark’s Peter Fenton, bringing total funding to $88 million. Launched in 2023, Ollama has become a go-to for programmers who want to get open-weight AI models up and running on a PC in minutes—no cloud required.
The company has racked up 176,000 GitHub stars and nearly 17,000 forks, and Morgan says it’s now used by over 8.9 million developers each month, including 85% of the Fortune 500—all with just 14 employees. Ollama also offers a neocloud service with subscription tiers from free to $100 per month, where users can access larger models and pay based on GPU time rather than token counts. Morgan and co-founder Michael Chiang previously built Docker Desktop, which was acquired by Docker after it bought their startup Kitematic. The parallel is clear: Ollama did for AI what Docker did for cloud computing—abstract away the hardware headaches.
Benchmark’s Fenton, who joined the board after the Series A, says the founders have a rare talent for building developer tools that achieve ubiquity. He noted that their work at Docker now serves over 10 million developers daily. The business case for Ollama sharpened in January, when the open model OpenClaw went viral and proved that open-weight models could handle agentic tasks like coding. Since then, enterprises and AI startups have been shifting more workloads to open models to cut inference costs—a trend that benefits Ollama’s cloud business. Fenton argues it’s not a zero-sum game between open and closed models, but says any company with high inference expenses has a strong incentive to move to open-weight models. Despite some community grumbling about the company monetizing through its cloud service, Morgan insists the core free desktop product remains unchanged, and the cloud offering simply extends the mission: helping developers find and use models that are too large to run locally.
Source: TechCrunch