Why 85% of the Fortune 500 already runs Ollama: Our Investment in their Series B

How do organizations move a local LLM from an engineer’s laptop to company-wide infrastructure? Michael Chiang and Jeff Morgan, co-founders of Ollama, answered that question by building a product that favors choice, private data, and company-owned models. After cutting their teeth at Docker Desktop, they brought their instinct for design that lets builders bring their own data, customize behavior, and keep IP in-house. 

This thesis is proven by Ollama’s scale:  almost 8.9 million active developers, nearly a million new installs per week, and an 85% Fortune 500 footprint, plus a hosted path for massive workloads. For these reasons and more, we’re excited to announce we’re leading Ollama’s $65M Series B.

In our conversation, we discuss:

• Nearly a million new installs a week, on top of almost 8.9 million active developers
• The Docker Desktop muscle behind the open-source instinct
• Why the open model pitch is about control, owned IP, customization, and data
• How experiments at home lead to 10-20 people using the product inside adopters’ companies
• NASA staffing, a Finnish power plant, and Lawrence Livermore’s particle accelerator as real deployment patterns
• The Cloud offering’s role hosting massive models like DeepSeek in the US and Europe
• Where the $65M goes: team, Cloud growth, and faster open-source performance on more hardware

‍

3 takeaways from this conversation:

1. Control is the wedge. Cost and privacy matter, but the deeper pull is that teams can build around their own IP, tune model behavior, and run on company-owned data without handing the whole workflow to someone else.

2. The dev loop starts on a laptop, then spreads through the company. An engineer tries Ollama at home, brings a working pattern into a meeting, and suddenly 10 or 20 people are using it internally. That bottom-up path is the enterprise motion.

3. Local and hosted are becoming one architecture. Ollama’s Cloud work extends local rather than replacing it; it supports massive models and regional capacity while keeping the same model choice, data-control, and deployment philosophy