Ollama Raises $65M as Local AI Models Go Mainstream

Ollama, the largest developer platform for running open-source AI models locally, has closed a $65 million Series B led by Theory Ventures. The raise, reported Monday, positions Ollama as the dominant infrastructure layer for developers and enterprises that want to run open-weight AI models without sending data to a third-party cloud.

The funding round underscores how local inference has moved from a hobbyist preference to a mainstream enterprise requirement.

The AI Insider report on the raise said Theory Ventures led the round with additional participation from existing backers. Specific valuation figures were not published alongside the announcement.

What Ollama Does and Why It Has Explosive Developer Adoption

Open-source AI models, sometimes called open-weight models, are model architectures and trained weights that researchers and companies release publicly.

Examples include Meta’s Llama family, Mistral’s models, and Google’s Gemma series. Unlike proprietary models from OpenAI or Anthropic, anyone can download and run them on their own hardware.

Running them is the hard part.

Open-weight models arrive as large files of numerical weights, sometimes exceeding 70 gigabytes, that require careful software configuration to load and run efficiently on consumer or enterprise hardware. Ollama abstracts all of that complexity into a single command-line tool.

A developer types one command, and the model downloads, configures, and runs locally.

That simplicity has driven adoption that rivals far better-funded competitors. Ollama’s GitHub repository has accumulated over 100,000 stars, making it one of the most-starred AI infrastructure projects in the open-source ecosystem.

The platform supports hundreds of models and runs on Mac, Windows, and Linux without GPU requirements, though GPU acceleration is supported when available.

The Privacy and Cost Logic Driving Enterprises Toward Local AI Models

The case for local inference splits into two arguments. The first is privacy.

Enterprises in healthcare, law, and finance handle data that cannot legally or contractually leave their own infrastructure. Sending queries to a cloud API, even an encrypted one, creates a data-transfer event that compliance teams often cannot approve.

The second argument is cost.

Cloud inference APIs charge per token, and at high query volumes, those costs compound quickly. A company processing millions of documents per month can spend more on API fees than on the salary of the engineers building the product.

Running capable models locally, once the hardware cost is amortized, can reduce per-query cost to near zero.

Both arguments have become more compelling in 2026 as open-weight model quality has closed the gap with proprietary alternatives. Meta’s latest Llama release performed within a few percentage points of GPT-4 on several standard benchmarks, and Mistral’s models have shown strong results on code and reasoning tasks at a fraction of the parameter count.

Theory Ventures’ Thesis and the Open-Source AI Models Stack Taking Shape

Theory Ventures, the fund led by Tomasz Tunguz, has built a portfolio concentrated on developer infrastructure.

The firm’s bet on Ollama fits a pattern of backing the lowest layer of the stack, the tools developers touch first when building an AI application.

The infrastructure thesis is straightforward. If open-weight models continue to improve and enterprise adoption of local inference grows, the platform that controls model discovery, download, and runtime becomes the default distribution channel for every open-weight release.

Ollama currently occupies that position.

The competitive risk is fragmentation. LM Studio, Jan, and several cloud providers offer local inference options that increasingly match Ollama’s feature set. The platform’s lead is currently based on simplicity and ecosystem momentum, both of which erode if a well-funded competitor ships a meaningfully better developer experience.

How the Ollama Round Connects to the Broader Open-Weight Moment

Ollama’s $65 million raise follows a stretch of months in which open-weight model releases have consistently surprised the market with their capability levels.

The AI models available in mid-2026 would have outperformed most closed APIs available in early 2024.

That pace of improvement has changed how enterprises plan AI infrastructure. A year ago, most enterprise AI architects assumed that frontier-quality inference would always require a cloud API.

That assumption is now contested. Ollama’s Series B is, in part, a bet that the assumption will prove wrong for a large and growing segment of the market.

The Infinity raise earlier on July 20 reflects the same underlying trend from a different angle.

Where Infinity targets enterprises that want to run inference through a managed service at lower cost, Ollama targets developers and enterprises that want to run AI models themselves. Both companies are betting that inference, not model training, is where the next layer of AI infrastructure value will be built and captured.

Read Next: Open-Weight AI Models Now Close the Dangerous Cyber Gap