Ollama raises $65M to connect 9 million developers to open-weight models, with 80% of Fortune 500 already using the platform

Ollama raises $65M to bring open-weight models into the enterprise

Ollama has raised $65 million, led by Tomasz Tunguz of Theory Ventures, with existing investors also participating. The company, built by a team of 14 people, describes itself as the largest network for developers to access open-weight AI models — either downloaded locally for low-latency edge use cases or hosted on US and European servers for teams with data residency requirements.

The headline traction figure is striking: 80% of Fortune 500 companies are already using the platform, driven almost entirely by bottom-up developer adoption. Individual developers bring Ollama to work for personal productivity, it spreads to their team, and at that point the conversation shifts from self-serve to a multiparty enterprise sale involving security, IT, and compliance teams. Morgan says 9 million developers are now on the platform.

“65,000,000. The lead was Tomasz Tunguz... We got here with 14 people to a company of this magnitude... GLM five two is the highest token volume of accessing [open models on Ollama], and I'm a daily GLM five two user through Ollama right now, and it's replaced 80% of my coding work.”

Open-weight models in the enterprise

Morgan's core argument is that open-weight models fit enterprise security requirements better than proprietary APIs because the weights can be deployed inside a company's own environment without exposing data to third-party infrastructure. For teams that do want cloud hosting, the data residency question is decisive: Morgan says a large share of open-model consumption today flows through servers with no data retention guarantees. Ollama's answer is US and European compute, which he frames as a hard requirement for most large enterprise customers — and a material use of the new capital.

The $65M goes toward two things: expanding the team to handle the enterprise sales motion (Ollama launched a Teams plan and was immediately inundated with thousands of signups), and building out that compute infrastructure.

Model preferences and the open-versus-frontier gap

Morgan says customers are using a mix of open and frontier models, and the choice is less about country of origin than about where the model runs and what safeguards are in place. He names GLM 5.2 as the current highest-volume model on the platform by a wide margin, and cites NeMo Tron 3 Ultra and Google DeepMind's Gemma as strong US-built alternatives. He says GLM 5.2 has replaced 80% of his own coding work.

On whether the gap between open and frontier models is closing or widening, Morgan says it oscillates but trends narrower overall. His more pointed claim is that customers fine-tuning open models on their own data are regularly outperforming generic frontier models on their specific tasks — and that this advantage will compound as agentic use cases mature.

As launch partner with every major model lab, Ollama positions itself as the connective layer between model releases and the developer community large enough to generate real-world signal on which models perform for which tasks.