Analysis
Ollama, the open-source platform that lets developers download and run large language models locally rather than through a cloud API, closed a $65 million Series B funding round led by Theory Ventures on July 9. Benchmark, 8VC and Y Combinator also participated, bringing the three-year-old company's total funding to $88 million.
The round lands on the back of rapid developer adoption: Ollama now counts 8.9 million monthly active developers, more than double the 4.45 million it had in January, with nearly 1 million weekly installs and a presence in 85% of Fortune 500 companies spanning government, healthcare and finance. The company operates with a lean team of just 14 people.
Ollama's model is to act as a distribution and hosting layer for the open-weight model ecosystem -- it hosts heavyweight open models including Nemotron, GLM, DeepSeek, Kimi and MiniMax, and partners with chipmakers Nvidia, AMD, Intel and Qualcomm to ensure users can run new model releases on day one regardless of underlying hardware. That positions the company less as a model developer itself and more as critical infrastructure for whichever open models win.
“That positions the company less as a model developer itself and more as critical infrastructure for whichever open models win.”
Founders Jeff Morgan (CEO) and Michael Chiang have run this developer-tooling playbook before: their previous company, Kitematic, was acquired by Docker in 2015 and became the foundation for Docker Desktop, a product now used by more than 10 million developers. Benchmark partner and new board member Peter Fenton framed the bet on open-weight models bluntly, saying open-weight models will generate the supermajority of tokens within the next 18 to 24 months.
For founders building on top of open models, Ollama's growth curve is a useful signal that the demand for local, self-hosted inference hasn't been crowded out by API-first providers -- if anything, enterprise deployments in regulated sectors appear to be accelerating it. For investors, the round is a bet that distribution and developer experience, not model training itself, is where durable value accrues in the open-weight stack.
The bear case: Ollama doesn't train its own models and is dependent on the continued vitality of the open-weight ecosystem it distributes -- if labs like Meta, DeepSeek or Alibaba pull back on releasing competitive open weights, Ollama's core value proposition erodes with them. What to watch next: whether Ollama's cloud hosting business for heavyweight models becomes a meaningful revenue line on its own, and whether the 8.9-million-developer base converts into paid enterprise contracts at a rate that justifies the new funding.
