Ollama Raises $65M Series B, Reports 8.9M Monthly Developers

The interesting number in Ollama's $65 million Series B, reported by TechCrunch, is not the funding, it is the 14 employees. Fourteen people, reportedly serving over 8.9 million monthly developers, sitting inside 85% of Fortune 500 companies. That is the leverage story venture capital has been hunting for in AI infrastructure, and Theory Ventures just paid $65 million for the seat next to it.

Ollama itself is a local runner for open-weight models, the piece of software that lets a developer spin up a model on a personal computer within minutes rather than wiring up an API. The round brings total funding to $88 million on top of a $15 million Series A led by Benchmark's Peter Fenton, who now sits on the board. Fenton's pitch, per the reporting, is that open and closed models will coexist, and that enterprises staring at closed-model API bills are treating cheaper open alternatives as a "vital existential project."

The commercial model is a tell. Ollama is keeping the desktop product free, with the founders emphasizing that "nothing has changed for the core product that's free on the desktop," and layering paid subscription tiers up to $100/month on top. Usage is billed by GPU time rather than tokens, a small design choice that squares neatly with agentic workloads, where a single task can run long and quiet in the background. Founders Jeff Morgan and Michael Chiang have run a version of this playbook before, having built Docker Desktop after Docker acquired their earlier startup Kitematic.

The honest caveat is that the reporting does not disclose revenue, a valuation on the round, or whether the 85% Fortune 500 figure reflects real production deployments versus curious developer downloads. It is also a crowded stack, and free-tier gravity could cap how much of that 8.9 million base ever converts to paid, especially as rival local runtimes chase the same developers.

What makes the bet coherent is timing. Morgan pinpoints an inflection around January, when larger open models became useful enough for "agentic tasks, like coding." If that read holds, Ollama's real market is not hobbyists on laptops anymore, it is the enterprise workload teams who would rather run their agents on their own GPUs than meter every token through someone else's API.