Ollama raises $65 million as its free local-model tool hits nearly 9 million monthly developers

Illustration: BCN

Ollama, the tool a lot of developers use to run open AI models on their own machines, has raised a $65 million Series B led by Theory Ventures. Founder and CEO Jeff Morgan confirmed the round to TechCrunch. It follows a $15 million Series A led by Benchmark's Peter Fenton and brings the total raised to $88 million. The company would not discuss its valuation or revenue.

That is a lot of money for a piece of software you can download for free, so it helps to be clear about what Ollama is and where the business actually lives.

What it does

Ollama launched in 2023 as a way to run open-weight models on a laptop without fighting the setup. You install it, type ollama run and a model name, and a few minutes later you are talking to a model that lives on your own hardware instead of someone else's server. That is the whole pitch, and the reason it caught on: no API key, no per-token meter, no data leaving the machine.

The lineage explains the polish. Morgan and co-founder Michael Chiang helped build Docker Desktop before this, having landed at Docker when it bought their earlier startup, Kitematic. Docker made it painless to move a containerized app between machines and clouds. Ollama does the same trick for models. Package the weights, hide the hardware configuration, get it running in one command. It is a good piece of engineering aimed at a real annoyance, and developers noticed: the project has about 176,000 stars and 17,000 forks on GitHub.

The numbers, and what they leave out

Morgan says Ollama is now "used by over 8.9 million developers every month, sitting in 85% of the Fortune 500 and growing like crazy," and that the company has done it with 14 employees. The headcount is the credible part. The rest are company-reported figures, and it helps to know what they measure. A count of monthly developers for a free local tool comes from installs and telemetry, not signed contracts, and "sitting in 85% of the Fortune 500" means someone at those companies ran it, not that those companies pay for it. The distinction matters, because usage and revenue are not the same thing, and Ollama declined to talk about revenue.

None of that makes the adoption fake. It is genuinely popular. It does mean the interesting question is the one the raise is meant to answer: how do you turn nine million people running free software into a business.

The cloud is where the money is, and where the friction is

The answer is a hosted service. Alongside the local tool, Ollama runs a cloud that hosts larger models than most laptops can handle, on subscriptions from free to $100 a month, billed by GPU time rather than token count. The logic is simple. The strongest open models are now too big to run at home, so Ollama offers to rent you the compute.

The trouble is that the local-first promise is exactly what made the tool popular, and a paid cloud pulls the other direction. About a year ago, some users said as much, complaining in blog posts and on Reddit and Hacker News that the cloud push was pulling focus from the free project. The word that got attached was "enshittification," the now-standard label for a good product slowly bending toward its business model.

Morgan frames the cloud as an extension of the same mission, not a break from it. The big open models are, as he put it, "too big to run on your own computer. So we said, 'Hey, let's help find the compute for that.'" Fenton, who joined the board with the Series A, drew the line more firmly: "Nothing has changed for the core product that's free on the desktop. There's zero change to the premise that this is the place you can discover and run local models."

That is the right thing to say and also the thing every company says right before the balance shifts. Worth watching, not yet worth alarm. The free tool is still free and still the reason anyone shows up.

Why an investor pays $65 million for this

The wager underneath the round is that open-weight models keep getting good enough to do real work, and that companies with large inference bills move toward them to cut costs. Morgan points to early this year, when larger open models became able to handle agentic tasks like coding, as the moment the business case firmed up. If that shift holds, whoever sits between developers and open models has a useful position, and Ollama already sits there for a lot of them.

Fenton's version of the bet is that open and closed models are not a winner-take-all fight. "I still think that this is the part that most of the debate gets wrong. It's not an either/or," he said, arguing that companies with high inference costs still have strong reason to move work onto open-weight models even while they keep using closed ones. That is a reasonable read of where things are going. It is also, conveniently, the read that makes Ollama valuable. Both can be true.

What Ollama has that most AI startups raising this kind of money do not is a product people already chose on its own merits, before anyone was selling them anything. The Series B is a bet that the company can charge for the next part without breaking the part that got it here.