What Happened

Ollama, the open-source tool that lets developers run AI models locally on their machines, has raised a $65 million Series B round led by Theory Ventures. Founder and CEO Jeff Morgan confirmed the raise to TechCrunch, noting that this follows a $15 million Series A led by Benchmark's Peter Fenton. The company has now raised $88 million in total funding.

The numbers are striking for a company with only 14 employees. Ollama reports 8.9 million monthly active developers, presence in 85% of Fortune 500 companies, 176,000 GitHub stars, and 17,000 forks. Revenue and valuation were not disclosed.

Founded in 2023 by Morgan and co-founder Michael Chiang — both ex-Docker engineers who joined Docker through its acquisition of their previous startup Kitematic — Ollama essentially does for AI models what Docker did for cloud applications: abstracts away the hardware configuration pain so developers can get models running in minutes.

The company offers a free desktop tool for running open-weight models locally and a paid cloud service for hosting larger models that are too big to run on personal machines. Cloud pricing ranges from free to $100/month, with usage billed on GPU time rather than token limits — a notable departure from how most inference providers price.

Morgan pointed to January 2026 as a turning point, when open models became capable enough for agentic tasks like coding. That shift drove demand for both local prototyping and hosted inference of larger open models, benefiting Ollama's cloud business.

Why It Matters

Ollama's raise is the latest signal that local-first AI development tooling is becoming infrastructure-grade. The 85% Fortune 500 penetration claim — if accurate — means Ollama has embedded itself into enterprise developer workflows the way Docker did before monetizing. That creates both a massive distribution advantage and a governance blind spot for IT leaders who may not realize how many of their developers are running AI models locally.

The funding also fits a broader pattern: capital is flowing aggressively into the open-weight inference layer. Together AI raised $800M at an $8.3B valuation earlier in July. RunPod raised $100M in June. But Ollama's positioning is distinct — it's not a pure neocloud. It's a developer tool company with a cloud business attached, leveraging enormous open-source mindshare (176K GitHub stars) as its acquisition channel.

The strategic tension is real, though. Ollama faced community backlash roughly a year ago when developers accused the company of 'enshittification' — degrading the free product to push paid cloud services. Morgan and Fenton both insist the core free desktop product hasn't changed. Whether that holds as the company scales with fresh capital and pressure to generate revenue remains an open question.

Who Is Affected

AI application developers and ML engineers are the primary audience. Many are already using Ollama for local model prototyping. The hosted cloud tier is a newer offering worth evaluating against existing inference providers.

Enterprise IT and security teams should take note of the 85% Fortune 500 claim. If developers are running open-weight models locally on company laptops, that has implications for data governance, model audit trails, and security policies.

Neocloud and inference providers — Together AI, RunPod, Modal, Replicate, and others — now face a well-funded competitor with massive developer distribution. Ollama's GPU-time-based pricing model is also a differentiator worth watching.

Strategic Implications

For AI startup founders: Ollama's local-first workflow is now the default developer experience for nearly 9 million developers working with open-weight models. If your product touches model inference or developer tooling, consider whether Ollama compatibility is a distribution lever. The company's 14-person team generating this scale suggests extreme capital efficiency — and a moat built on developer habit, not infrastructure.

For developers/operators building with AI APIs: Ollama's GPU-time-based pricing (vs. token limits) is a fundamentally different cost model. If you're prototyping locally with Ollama and then deploying to a cloud inference provider, evaluate whether Ollama's hosted tier reduces friction or cost in that transition. The ability to move from local to hosted without changing tooling is a real workflow advantage.

For non-technical business owners evaluating AI tools: Your engineering team may already be using Ollama. Ask your IT leadership about governance policies for local model usage — particularly whether sensitive company data is being processed on developer machines running open-weight models outside your approved vendor stack.

What to Watch Next

Monitor whether Ollama discloses revenue figures or valuation in subsequent filings, and whether the company expands its hosted cloud tier to compete more directly with Together AI and RunPod. Also watch for enterprise governance responses — if 85% of Fortune 500 companies have developers running local AI models, expect CISO and compliance tooling to target this gap.

Frequently Asked Questions

Q: What is Ollama and what does it do?

A: Ollama is an open-source tool that lets developers run open-weight AI models locally on their computers, getting them running in minutes without complex hardware configuration. It also offers a paid cloud service for hosting larger models that can't run on personal machines.

Q: How much did Ollama raise and who led the round?

A: Ollama raised $65 million in a Series B round led by Theory Ventures. This follows a $15 million Series A led by Benchmark's Peter Fenton, bringing total funding to $88 million.

Q: How many developers use Ollama?

A: According to the company, Ollama is used by over 8.9 million developers monthly and is present in 85% of Fortune 500 companies, maintained by a team of just 14 employees.

Q: Is Ollama free to use?

A: The core desktop tool for running models locally is free and open source. Ollama also offers a cloud hosting service with tiers ranging from free to $100/month, with pricing based on GPU time rather than token limits.