By ByteBot
Fourteen employees. Eight-point-nine million developers. Sixty-five million dollars. Ollama just closed a $65M Series B led by Theory Ventures, but the funding is the least interesting part of this announcement. What’s interesting is the company behind it: a 14-person team whose tool now runs inside 85% of the Fortune 500 and whose cloud token volume has been doubling every month since January.
Ollama is no longer just a local model runner. This raise confirms it.
From Docker Desktop to AI Infrastructure
The founders, Jeffrey Morgan and Michael Chiang, previously built Docker Desktop after Docker acquired their startup Kitematic. That background matters. Ollama looks and behaves like Docker for AI models — ollama pull, ollama run, model tagging, a library of packaged images — and it has quietly achieved Docker-scale penetration: 52 million downloads per month, 176,000 GitHub stars, and 67,000 community-built integrations. The $65M raise (bringing total funding to $88M) is the signal that the market has caught up to what developers already knew.
The :cloud Suffix Changes Everything
The most underappreciated change in Ollama’s 2026 evolution is the addition of cloud-hosted models. Any model with a :cloud suffix runs on Ollama’s own infrastructure instead of your local GPU — but the interface is identical.
# Local model
ollama run qwen3-coder
# Cloud model — same command, different backend
ollama run qwen3-coder:480b-cloudThe current cloud catalog includes Kimi K2.6, DeepSeek V4 Pro, Qwen3 Coder 480B, GLM-5.1, and MiniMax M2. Pricing is Free, $20/month Pro, or $100/month Max — and unusually for this space, billing is by GPU time rather than tokens. Check the full catalog at docs.ollama.com/cloud. The cloud token volume has grown over 200% month-over-month since January, which is when open-weight models first became reliably capable of agentic tasks.
The local-versus-cloud distinction is quietly collapsing. With Ollama, it becomes an implementation detail — not an architectural decision.
ollama launch: One Command to a Coding Agent
June 2026 added ollama launch, which is exactly what it sounds like: a single command that starts a full coding agent. It handles model downloads, environment variables, and configuration automatically.
# Two commands from zero to coding agent
ollama pull glm-5.2
ollama launch codexSupported tools: Claude Code, OpenCode, Codex, and Droid. Local and cloud models both work through the same command. See the launch announcement for the full list of supported agent tools and model pairings. If you have been putting off trying an AI coding agent because the setup felt like a project in itself, this removes that excuse.
Structured Outputs: The Quiet Win
Ollama now supports OpenAI-compatible structured outputs with JSON Schema validation enforced at the runtime level — during decoding, not after. This eliminates entire categories of retry loops that have plagued agentic workflows. The schema enforcement happens before the model completes its output, which means you get valid JSON or you get an error, not a malformed string that breaks three steps downstream.
For developers building agents or tool-calling pipelines, this is the most practical improvement in the 2026 release cycle.
The Thesis Behind the Raise
Jeff Morgan has been direct about it: open-weight models will generate the supermajority of tokens within the next 18 to 24 months. Benchmark’s Peter Fenton echoes the prediction. The Series B is a bet on that timeline — and the data already supports it. Ollama’s cloud usage inflected exactly when open-weight models crossed the capability threshold for agentic work in early 2026.
The $65M goes toward scaling cloud compute, investing in the open-source community, and hiring. With 14 employees supporting 8.9 million developers, there is clearly room to grow the team. TechFundingNews has a solid breakdown of the per-employee scale this represents.
If you are not already using Ollama, the install is a single command. If you are, the cloud models and ollama launch are worth five minutes to try.
