Jeff Morgan's Ollama raises $65M as local AI becomes a cloud business

Why it matters

Ollama shows how open-source AI infrastructure is monetizing: win local developer workflow first, then sell cloud compute when agents and larger models outgrow laptops.

Jeff Morgan and Michael Chiang's Ollama has raised a $65 million Series B led by Theory Ventures, Morgan told TechCrunch in a July 9 report, putting fresh venture money behind one of the default ways developers run open-weight AI models on their own machines.

The round brings Ollama's total funding to $88 million. It follows a $15 million Series A led by Peter Fenton at Benchmark, who joined Ollama's board, according to TechCrunch. Ollama did not disclose the Series B valuation, revenue, or any participating investors beyond Theory Ventures.

The founder story matters here because Ollama is a second run at a problem Morgan and Chiang have already lived through. Before Ollama, they built Kitematic, sold it to Docker, and then helped build Docker Desktop. Their earlier work made container workflows usable for everyday developers; Ollama applies the same instinct to open models, where the first wave of releases in 2023 was powerful enough to be interesting and awkward enough to repel most programmers. Morgan told TechCrunch that open models were "really hard to use" when they started arriving in 2023.

Ollama's first wedge was simple: install the tool, pull a model, run it locally. The Ollama GitHub repo lists about 176,000 stars and 16,900 forks, a strong public signal for developer adoption, though GitHub stars measure attention and intent rather than active usage. Morgan told TechCrunch that Ollama is used by more than 8.9 million developers monthly and is present in 85% of the Fortune 500. Those figures are founder-provided and were not independently broken down by TechCrunch.

Y Combinator's company directory lists Ollama as a Winter 2021 company in artificial intelligence, developer tools, and open source, with Morgan and Chiang as active founders. TechCrunch says Ollama launched in 2023, which means the product's public adoption curve has compressed into roughly three years.

The local tool is turning into a cloud meter

Ollama's Series B is less about a free desktop utility and more about whether Morgan and Chiang can convert that developer habit into paid compute.

Ollama's site still leads with local setup: a terminal command, a CLI, and the promise that developers can build with open models on their own hardware. The same homepage now pushes a second motion: "Start local. Scale with cloud." Ollama says its cloud gives users access to larger models on datacenter-grade hardware, parallel requests, and web-connected model runs when local machines hit memory or performance limits.

That cloud layer is where Ollama's business model becomes visible. The pricing page lists a free tier, a Pro plan at $20 per month or $200 per year, and a Max plan at $100 per month. Pro includes three cloud models at a time and 50x more cloud usage than Free. Max includes 10 concurrent cloud models and 5x more usage than Pro. Ollama says usage is based primarily on GPU time rather than a fixed token or request cap.

That pricing choice is important. Token-based plans are easy to compare across API providers but can be hard for users to map to actual infrastructure cost. GPU-time pricing ties the product closer to compute scarcity, which is the constraint Ollama is trying to abstract away. It also gives Ollama room to change the economics as model architectures and hardware improve.

The product shift also explains why the round happened now. Morgan told TechCrunch that Ollama's business case became clearer around January, when larger open models became useful for coding and other agentic tasks. Ollama's own blog tracks that shift: in January, Ollama added OpenAI Codex CLI support and Anthropic Messages API compatibility; in February, it added Claude Code subagents and web search; by September 2025, Ollama had introduced cloud models in preview.

RuntimeWire reported in June that Ollama was already pushing the other side of the same thesis: quantized weights that could make a claimed Gemma 4 31B model run on consumer laptop hardware. The Series B says investors believe both tracks can compound together. Better local inference keeps Ollama credible with developers who care about control. Cloud inference gives Ollama a revenue line when the same developers want bigger models, longer agent runs, or fewer hardware limits.

The Docker comparison cuts both ways

TechCrunch frames Ollama as an AI-era parallel to Docker, and the comparison is fair as far as developer experience goes. Docker turned containers from an infrastructure concept into a daily workflow. Ollama took open-weight models, which had been scattered across model hubs, papers, repos, and hardware guides, and turned them into a command developers could remember.

The comparison also exposes the pressure Ollama faces. Docker became foundational infrastructure, then had to build a commercial model around a developer community that expected the core workflow to remain free and broadly available. Ollama is reaching that same point earlier in its life. The free local product gives Ollama distribution. The cloud product pays for the large-model experience that local hardware cannot always handle.

Morgan's answer, per TechCrunch, is that larger open models often exceed what local machines can handle, so Ollama is helping users find the compute.

That is the line Ollama has to hold. Open-source developer tools can monetize if users believe the paid product expands the core project rather than replacing it. The risk is visible in public developer forums and issues, where some users have complained that Ollama's commercial cloud work could pull attention away from the free local project. A separate GitHub issue asked Ollama to clarify licensing for the newer Ollama app compared with the MIT-licensed core repo. The issue is a narrow licensing question, but it reflects a broader expectation: developers want to know where the open-source boundary sits before they build workflows around a tool.

A crowded market, with one advantage

Ollama does not own local inference. LM Studio offers a desktop-first way to download and run local models, serve them through OpenAI-like endpoints, and use llama.cpp or Apple's MLX. Jan positions itself around offline private chat on desktop. LocalAI targets server-side users with a local, OpenAI-compatible API surface. Docker itself made Docker Model Runner generally available in September 2025, bringing local LLM management into Docker Desktop and Docker Engine.

Ollama's advantage is habit. A tool that starts as a developer's first local model runner can become the default path for model discovery, coding-agent setup, cloud spillover, and eventually team governance. Ollama's homepage already lists support for running apps and agents such as OpenClaw and Claude Code-style workflows. The repo's README lists integrations across code editors, terminals, app frameworks, chat interfaces, and observability tools.

The open question is whether that usage converts into durable paid accounts. Ollama has not disclosed revenue. It has not disclosed valuation. The 8.9 million monthly developer figure, the 85% Fortune 500 claim, and the 14-person headcount all come from Morgan through TechCrunch. Those numbers describe remarkable reach, but they do not reveal retention, paid conversion, cloud gross margins, or how much usage is still purely local and free.

There is also a security side to Ollama's success. In January 2026, SentinelLABS and Censys reported finding 175,108 publicly reachable Ollama hosts across 130 countries during a scanning study. Ollama binds to localhost by default, according to that report, but users can expose it publicly through configuration. That is the cost of a tool becoming infrastructure: misconfigured deployments become part of the story.

The Series B gives Morgan and Chiang room to build through that stage. Theory Ventures, a San Francisco firm that says it invests at inception through Series B in software companies using technology shifts as go-to-market advantages, is backing Ollama at the moment when open models are moving from experimentation into daily coding work. Benchmark's earlier bet was on the founders' product judgment. Theory's bet is that the distribution they created can support a cloud business without burning the trust that made Ollama useful in the first place.