Popular open source AI developer tool Ollama raises $65M, grows to nearly 9M users

The fashionable unfastened root AI instrumentality Ollama has raised a $65 cardinal Series B, led by Theory Venture, laminitis and CEO Jeff Morgan tells TechCrunch.  

This circular follows a erstwhile $15 cardinal Series A led by Benchmark’s Peter Fenton. All told, the institution has present raised $88 million.  

Ollama, which launched successful 2023, helps devs tally open-weight AI models connected their PCs, getting them up and moving successful minutes. It has been praised by developers crossed countless training sites, videos, blogs and social media posts. It has amassed 176,000 stars and astir 17,000 forks on GitHub.  

Developers tin besides usage Ollama to find models and entree larger, much analyzable ones that it hosts connected its neocloud via respective subscription tiers, from escaped to $100/month. It besides tracks usage based connected GPU time, not token limits. 

If the ngo to assistance developers much easy physique connected their PCs sounds vaguely familiar, it should. Morgan and his co-founder Michael Chiang antecedently helped physique Docker Desktop. They landed astatine Docker aft it bought their erstwhile startup, Kitematic. Docker makes containers that assistance unreality apps casual to determination from unreality to cloud, oregon from desktop to cloud, abstracting distant each the pesky hardware configuration issues.

So Ollama fundamentally did for AI what Docker and Docker Desktop did for cloud. 

“Open models started coming retired successful 2023 but they were truly hard to use,” Morgan said. They had been geared toward researchers astatine the time, not programmers. “As a result, it was truly hard to get them up and running.” Three years aft launching, Ollama is present “used by implicit 8.9 cardinal developers each month, sitting successful 85% of the Fortune 500 and increasing similar crazy,” helium said. All with lone 14 employees.

That vocation acquisition is what drew Benchmark’s Peter Fenton to pb its earlier circular and articulation the board.

“What Jeff and Michael built with Docker is being utilized by 10 million-plus developers each day. The originative powers to make a merchandise that goes to ubiquity for developers is highly rare,” Fenton told TechCrunch. 

Morgan and Fenton declined to sermon the startup’s revenues and caller valuation. However, Morgan says that the proving constituent for Ollama arsenic a concern happened astir January, erstwhile OpenClaw became hot. That’s erstwhile larger unfastened models “suddenly became capable to bash these agentic tasks, similar coding. Obviously, we saw the detonation of the assistants similar OpenClaw, and this thought that unfastened models tin get existent enactment done.” 

Since then, the manufacture has been abuzz with the thought that paying users (particularly deep-pocketed enterprises and fast-growing AI application-layer startups) volition progressively crook to much affordable unfastened models, reserving their usage of closed models similar Anthropic for much of an as-needed basis. 

“I inactive deliberation that this is the portion that astir of the statement gets wrong. It’s not an either/or,” Fenton says of unfastened versus closed AI models. There volition beryllium plentifulness of concern for both, helium contends. However, each institution with precocious inference expenses  — the costs of utilizing the models  — has a “vital existential project” pushing them to determination “to open-weight models,” helium says.

There’s plentifulness of grounds that specified startups and enterprises are already turning to unfastened models for their regular needs. That, obviously, bodes good for Ollama’s unreality business.

But adjacent much interesting, Ollama is different illustration of however AI is birthing a ample caller harvest of unfastened root projects that are turning into companies pursued by VCs. There are unfastened root inference providers similar Inferact, shaper of vLLM, and RadixArk, shaper of SGLang. There is OpenClaw and its alternatives like NanoClaw. There are adjacent tiny startups gathering their ain unfastened models from scratch, similar Arcee.

To beryllium sure, not each Ollama instrumentality has been blessed that the institution has been pursuing making a living. About a twelvemonth ago, a bunch of blog and social media posts complained that its unreality concern was drafting attraction distant from its beloved escaped task and cited Ollama arsenic an illustration of the so-called “Enshittification” of dev tools, arsenic the inclination is called.  

But Morgan sees its unreality work arsenic an improvement of its unfastened root ngo to assistance programmers find and easy usage models. Those state-of-the-art, large, unfastened models are often “too large to tally connected your ain computer. So we said, ‘Hey, let’s assistance find the compute for that,’” helium explained. 

Board subordinate Fenton adds, “Nothing has changed for the halfway merchandise that’s escaped connected the desktop. There’s zero alteration to the premise that this is the spot you tin observe and tally section models.”