Ollama, a two-founder startup known for doing to local AI what Docker did to cloud containers, closed a $65 million Series B on July 9, 2026, led by Theory Ventures. Total funding now stands at $88 million. A project that started as a niche tool for researchers has grown into daily infrastructure for nearly 9 million developers per month.
Key takeaways
- Series B: $65M led by Theory Ventures
- Total funding: $88M (prior: $15M Series A from Benchmark Partners)
- Monthly active users: 8.9 million developers
- Present in 85% of Fortune 500 companies
- GitHub: 176,000 stars, nearly 17,000 forks
Docker for local AI models
Jeff Morgan and Michael Chiang, Ollama's founders, have a well-documented track record in developer tooling: both helped build Docker Desktop and joined Docker through their earlier startup Kitematic. That context explains why their approach to local AI prioritized launch simplicity over model parameters from the start.
When open-weight models began leaving research environments in 2023, they were hard to use — requiring CUDA environment configuration, manual dependency management, and understanding quantization formats. Ollama reduced that to a single terminal command and quickly became the default test environment for anyone running Llama, Mistral, or Gemma locally without sending data to an external API.
What Jeff and Michael built with Docker is used by 10 million-plus developers every day. The creative powers to create a product that goes to ubiquity for developers is extremely rare.
Peter Fenton, Benchmark Partners — investor and board member.
Business: not just free software
Ollama builds revenue on two mutually reinforcing layers. The first is the free desktop app — still being developed with no licensing changes, responsible for the user base. The second is a cloud service where Ollama hosts larger models requiring datacenter-grade hardware. Billing is based on GPU time, not token limits — which in practice means more predictable costs for development teams.
The business model's turning point came in January 2026, when agentic coding tools based on open-weight models took off. As Morgan explains, large open models "suddenly became able to do agentic tasks, like coding. Obviously, we saw the explosion of the assistants like OpenClaw, and this idea that open models can get real work done." This shifted users from experimentation to production — and triggered demand for Ollama's cloud layer.
Open-source AI ecosystem as a new startup class
Ollama's growth illustrates a broader trend: open-source projects that were previously funded through grants or sponsoring companies are now building standalone venture-backed companies. In the same segment are Inferact (maker of vLLM) and RadixArk (SGLang) — both developing open-source inference engines for large language models.
Fenton views this trend optimistically: "It's not an either/or" on open versus closed models. Every company with high inference costs has, in his view, a "vital existential project" to migrate daily operations to open models while maintaining access to closed ones — like Anthropic models — for tasks requiring frontier capabilities.
Not all Ollama fans are happy with the project's commercialization. About a year ago, discussion platforms saw pushback calling the new subscription plans "enshittification" of developer tools. Morgan defends the commercial layer as a natural extension of the mission: the largest models are "too big to run on your own computer," so Ollama helps find the right compute.
Why this matters
A $65M round for a 14-person startup with 8.9 million monthly active users demonstrates that local AI infrastructure has concrete business value — independent of debates about whether frontier models will remain closed. Ollama occupies the position between open-weight model producers (Meta, Mistral, Google) and end users, offering something simple: run what you want, where you want, without sending data outside.
For enterprises, that's a compliance and cost-control argument at the same time. Data stays in the client's infrastructure, costs are predictable, and dependency on a single API provider disappears. This explains why a tool present in 85% of the Fortune 500 attracted Series B interest despite its primary product being and remaining free.
For the open-source model ecosystem, it signals that the distribution layer — not just the models themselves — has standalone value. Tools that make open-weight AI easy to deploy can build durable businesses even without controlling the model weights.
What's next
- Ollama has announced cloud layer expansion — access to larger models from additional geographic regions (US, Europe, and Singapore are already available per the product page).
- Accelerated enterprise migration to open-weight models for everyday inference — a tendency Fenton named directly as a business driver — will increase traffic on the cloud platform.
- Competition in the open-source inference segment (vLLM, SGLang, LM Studio) will force continued investment in Developer Experience and model launch speed.
