Ollama builds software that lets developers run open-weight AI models on their own computers. It also lets them access larger hosted models through the same interface. In the latest Ollama funding round, the startup raised $65 million in Series B financing led by Theory Ventures. The new investment brings its total funding to $88 million. The bet is simple. Open models improved quickly, but they remain difficult to use because of setup steps, hardware challenges, and API complexity. Jeff Morgan and Michael Chiang founded Ollama in 2023. Morgan says the platform now reaches more than 8.9 million developers each month and is used by 85% of the Fortune 500.
What is Ollama and how does it work?
Ollama is basically a runtime and distribution layer for open-weight models. A developer installs it on macOS, Windows, or Linux, opens a terminal, and can start a local chat with a model through a single command like ollama run gemma4. The same workflow also works for hosted models. Users can invoke them with a cloud tag instead of switching tools or rewriting their application flow.
That's the big product trick. Ollama takes a pile of low-level tasks that used to be annoying — model downloads, local serving, API access, model naming, and environment setup — and turns them into a uniform developer experience. Its API supports completions and chat. It also handles embeddings, model pulls, pushes, copies, and local model inspection through a localhost server that developers can plug into apps and scripts.
It's also not just a chat wrapper. Ollama supports structured outputs, so developers can force model replies into JSON or a defined schema, which makes it much easier to build apps that need predictable results instead of fuzzy prose. It also supports tool calling, including parallel tool calls and multi-turn agent loops. That matters.
And for teams that want more control, Ollama exposes a Modelfile format that acts like a recipe for custom model behavior. Developers can build from an existing model, import supported Safetensors or GGUF weights, set context windows and temperatures, add system prompts, and apply LoRA adapters. That's a lot cleaner than stitching together separate model files, inference backends, and prompt templates by hand.
Who founded Ollama and what traction does it have?
The founding story
Morgan and Chiang didn't come into this cold. Before Ollama, they were part of the Kitematic team, and Kitematic was built to make Docker usable for normal developers instead of only the terminal die-hards. Docker acquired Kitematic in March 2015, describing it as the fastest and easiest way to use Docker on a Mac, with one-click container setup and a graphical interface on top.
That history matters because Ollama is chasing a very similar wedge. In 2023, open models were getting better, but Morgan said they were still “really hard to use.” So the company tried to do for local and open-weight AI what Docker Desktop did for containers: hide the ugly plumbing without taking power away from developers.
Why Morgan and Chiang fit this market
The Kitematic story started when Jeff Morgan, Michael Chiang, and Sean Li were still living near the University of Waterloo and trying to simplify how developers handled modern application complexity. That project sold quickly, and the founders moved into Docker, where Morgan and Chiang helped build the experience layer many developers now associate with Docker Desktop.
That's why Benchmark's Peter Fenton got involved early. His logic is easy to understand: very few founders have already built a developer tool that became close to default infrastructure. Ollama isn't backed because local AI is trendy. It's backed because its founders have already shown they know how to turn painful setup into habit-forming software.
Traction, pricing, and fundraising
The numbers are the loud part. Ollama is now used by more than 8.9 million developers every month, is present inside 85% of the Fortune 500, and has done that with a team of only 14 employees. The open-source project has also built a massive GitHub footprint, with 176,000 stars and nearly 17,000 forks.
Beyond the free desktop app, Ollama makes money through hosted access to larger models on its neocloud. Subscription tiers range from free to $100 a month, and usage is tracked by GPU time rather than token limits. That's a pretty pointed contrast with mainstream model APIs, where costs can become hard to predict once usage spikes.
Theory Ventures led this new round, adding $65 million in Series B capital after a $15 million Series A led by Benchmark's Peter Fenton and bringing total funding to $88 million. Morgan and Fenton didn't disclose revenue or valuation. Morgan did say the business really started to click around January, when larger open models became capable enough to handle more agentic coding-style work.
How does Ollama compare with LM Studio, Jan, GPT4All, and vLLM?
The closest direct alternatives for individual developers are local AI desktop tools like LM Studio, Jan, and GPT4All. LM Studio focuses on local model experimentation and OpenAI-style local APIs. Jan pushes an open-source desktop assistant with offline use, agents, and MCP-style connectors. GPT4All leans hard into local AI and data sovereignty.
Ollama's edge is that it feels more like infrastructure than a chatbot app, but less intimidating than a pure serving engine. It gives developers a CLI and a local API. There's also a model library and a path from laptop use to hosted models without forcing a platform jump.
Then there's vLLM, which is a different animal. vLLM is a high-throughput, memory-efficient serving engine built for production inference and scale, not primarily for the “get this running on my machine in 5 minutes” crowd. Ollama's real incumbent isn't one named rival. It's the old mess of manual model downloads, incompatible formats, custom wrappers, and too many sharp edges.
Why does Ollama funding matter now?
This round matters because Ollama is trying to balance 2 businesses that usually pull against each other: beloved free developer tooling and paid hosted infrastructure. The desktop app made the brand. The cloud offering could make the economics work.
And the timing isn't random. Morgan's point about January is important. Open models stopped being just something developers tested on weekends and started becoming useful for coding assistants and agent-style workflows that can do real work. Once that happened, a tool that standardizes local and hosted model usage got a lot more valuable.
Fenton's thesis is blunt. Companies with large inference bills have a “vital existential project” to push more workloads toward open-weight models. He's also not buying the lazy open-vs-closed framing. His view is that most serious buyers will use both — paying for proprietary models when they need the best frontier output, but shifting everyday workloads to cheaper open alternatives when the trade-off makes sense.
That's why this Ollama funding round feels bigger than a routine dev-tools raise. It's a bet that the interface layer around open models can become strategic infrastructure, not just a handy open-source utility.
How big is the market for open-source AI tools?
The raw spend is already massive. Gartner forecast worldwide generative AI spending at $643.9 billion in 2025, up 76.4% from 2024, with hardware soaking up most of that total as AI-capable devices and servers spread through the market.
But the more relevant signal for Ollama is behavior, not just budget. GitHub said in early 2025 that in its survey of 2,000 enterprise respondents across the US, Germany, India, and Brazil, nearly everyone had experimented with open-source AI models. That doesn't mean everyone is standardized on them. It does mean the trial phase is already broad.
McKinsey's 2025 survey of more than 700 technology leaders and senior developers across 41 countries points the same way. More than 50% of respondents said their organizations were using open-source AI technologies across parts of the stack, and 76% expected their organizations to increase that usage over the next several years. Respondents also cited lower implementation costs and lower maintenance costs versus proprietary tools.
That doesn't kill closed models. It reinforces the hybrid future Ollama's backers are talking about. If teams want a mix of local, private, cheap, customizable, and occasionally very large hosted models, they need a layer that keeps the experience coherent. That's the lane Ollama is trying to own.
Final take on Ollama funding
A lot of AI startups are selling magic. Ollama is selling convenience — and honestly, that may be the sturdier business.
The product became popular because it removed friction right when open models became worth the trouble. Now the question is whether Ollama funding helps it turn that developer love into a durable cloud business without losing the community that made it matter in the first place. Two things to watch: whether the hosted side grows fast enough to justify venture money, and whether a 14-person team can keep shipping before bigger platforms copy the same playbook.
Read how Hakimo raised a $12M growth round led by Zigg Capital to expand its AI-powered physical security platform that transforms existing surveillance cameras into real-time monitoring and intelligent threat detection systems.
FAQ
- What is the latest Ollama funding round?
Ollama has raised a $65 million Series B led by Theory Ventures. That comes after a $15 million Series A led by Benchmark's Peter Fenton, bringing the company's total funding to $88 million. - How does Ollama work for developers?
Ollama gives developers a local runtime and API for open-weight models, plus access to larger hosted models through the same workflow. A user can install it on macOS, Windows, or Linux, run a model from the terminal, and call it through a localhost API. It also supports tool calling, embeddings, and JSON-structured outputs. - Who founded Ollama?
Ollama was started by Jeff Morgan and Michael Chiang, the founders behind Kitematic, which Docker acquired in March 2015. Their earlier work focused on making container tooling easier for everyday developers, which is basically the same user-experience problem they're now attacking in AI. - What market is Ollama in?
Ollama sits in the open-source AI infrastructure and developer-tools market, especially the slice focused on local model execution and simplified inference workflows. It overlaps with desktop tools like LM Studio, Jan, and GPT4All. It also brushes up against serving engines like vLLM when teams move from experimentation toward production.
