Ollama’s $65 Million Raise Puts Its 14-Person Team to the Test

Ollama raised $65 million this month to keep developers running AI models on their own hardware instead of paying for someone else’s cloud. The Series B, led by Theory Ventures, pushes the company’s total funding to $88 million. A tiny in-house team built and still runs the entire platform behind it.

That platform now sits inside most of the world’s largest companies. A year before this round closed, developers who helped build its audience accused the company of drifting from its free, open roots toward a paid cloud tier. The growth is real. So is the doubt that came with it.

Theory Ventures Leads the $65 Million Round

Ollama lets a developer type one command and run an open-weight AI model locally on a laptop, then switch to a bigger model in Ollama’s cloud without changing any code. The round closed on July 9, 2026, with Theory Ventures as lead investor.

Existing backers Benchmark, 8VC, Y Combinator, Pace Capital, 49 Palms and GTMFund all returned. Angel investors joined too, including Docker co-founder Solomon Hykes, Aaron Katz and Spencer Kimball.

Ollama named seven backers joining the round in its funding announcement without disclosing a valuation. Founder and CEO Jeffrey Morgan and Benchmark general partner Peter Fenton both declined to discuss revenue or the new valuation with reporters.

Fenton led that earlier $15 million Series A and joined Ollama’s board, where he still sits. Fenton has watched this pattern before. “What Jeff and Michael built with Docker is being used by 10 million-plus developers every day,” he told TechCrunch. “The creative powers to create a product that goes to ubiquity for developers is extremely rare.”

Jeffrey Morgan and co-founder Michael Chiang started Ollama in 2023, three years before this raise. Before that, they built Kitematic, a company Docker acquired in 2015, and went on to build Docker Desktop.

The Fourteen-Person Fortune 500

Ollama’s own numbers show the gap between headcount and reach.

• 8.9 million monthly active developers, roughly double the count Ollama reported in January.
• 67,000+ community-built integrations connecting the platform to coding agents and assistants.
• 85% of Fortune 500 companies now run Ollama somewhere in their stack, including government, healthcare and finance customers.
• 14 employees built and still maintain the entire platform.

The open-source repository backs up the claim. Ollama has collected 176,000 stars and nearly 17,000 forks on GitHub, placing it among the most starred AI infrastructure projects on the platform.

Why Founders Are Ditching the Per-Token Bill

Cloud AI bills scale with usage, and agentic workloads can run long in the background, quietly racking up per-token charges. Running an open model on hardware a company already owns caps that cost at the price of the hardware itself.

Morgan says the clearest proof arrived in January 2026, when OpenClaw, an autonomous AI agent tool, took off and pushed larger open models into daily coding and agent work. “Open-weight models will generate the supermajority of tokens within the next 18 to 24 months,” said Peter Fenton, the Benchmark general partner who sits on Ollama’s board.

As open models close the gap for most real work, the platform where AI runs becomes one of the most valuable positions in software.

Tomasz Tunguz, general partner at Theory Ventures and the investor who led this round, wrote that line announcing the firm’s bet.

Ollama’s edge is distribution as much as software. The company is a day-zero access partner for model labs including Meta, Google DeepMind, Mistral and MiniMax, and for hardware makers NVIDIA, Intel, AMD and Qualcomm. New model releases show up inside Ollama before many competing tools can package them.

The broader market backs the bet. The AI inference market was worth roughly $97 billion in 2024 and is projected to reach $254 billion by 2030, growing at an annual rate of about 17.5%, according to Grand View Research.

Who’s Really Running on Ollama?

Enterprise use goes well past coding experiments. Theory Ventures, the firm leading this round, says it already found deployments handling real operational work inside regulated industries, not side projects.

Tomasz Tunguz described monitoring electricity loads at a Finnish power plant, alongside deployments at a U.S. space agency, inside a public company’s finance office validating financial statements, and at a particle accelerator.

Ollama frames that spread as proof of its privacy pitch. When a model runs on hardware the customer already owns, sensitive data never has to leave the building.

The Enshittification Complaint That Won’t Go Away

About a year before the Series B closed, developer forums reached for a specific label. Cory Doctorow’s “enshittification” framework describes platforms that attract users, squeeze them to win business customers, then extract from both. Threads on Hacker News and Reddit pinned that label on Ollama’s cloud pivot.

Tech Times, reviewing the complaints around the raise, pointed to three specific incidents that fed the criticism.

• In July 2025, Ollama’s desktop GUI app shipped without an open source license, a gap the company corrected that November.
• The model library listed smaller, distilled versions of DeepSeek-R1 under the same name as the full 671-billion-parameter model, confusing users about which one they were actually running.
• A cross-domain token exposure vulnerability, tracked as CVE-2025-51471 with a CVSS severity score of 6.9, was reported and patched.

The criticism reached beyond licensing and naming. A separate controversy, dated to September 2025, saw developers say Ollama’s new web search API lacked clear privacy documentation for how it handled prompts and queries, a complaint the company answered by pointing to a zero data retention policy.

Morgan and Fenton reject the enshittification label outright. Their position is that the free desktop tool has not changed and that the cloud tier exists only to reach models too large for a personal laptop.

Can Ollama Turn Free Users Into Paying Customers?

Ollama has not disclosed revenue, a valuation for this round, or a specific new headcount target. That leaves one question hanging over the raise: whether a mostly free base of 8.9 million developers converts into paying cloud customers before a rival local runtime gets there first.

The company prices its cloud tier from free up to $100 a month and bills by GPU time rather than by the token, a structure suited to agentic tasks that can run unattended for long stretches. It competes for that conversion against rivals that have raised far more capital.

Ollama’s total funding is a fraction of what Hugging Face and Together AI have each raised on their own, even as Ollama calls itself the largest developer network among them. Leaner funding has not slowed its developer count, but it does mean less cash on hand if the conversion from free to paid takes longer than investors expect.

The Free Desktop Tool Stays Free, for Now

Ollama says the new capital goes toward cloud compute capacity, the open-source community, and new hires, though it has not set a specific headcount target beyond the current 14. The plan is to keep offering day-zero access to new open model releases as its main pitch to developers who don’t want to wait.

The one-line install that started this in 2023 still works exactly the way it always did. Type a single command, and a model runs on the machine in front of you, no sign-in required.

Frequently Asked Questions

What does Ollama actually do?

Ollama lets developers install and run open-weight AI models with a single command on a Mac, Windows or Linux machine. The same command and the same OpenAI-compatible API call can switch from a small local model to a much larger cloud-hosted one without changing any code.

Is Ollama still free to use?

The desktop app and local inference remain free. Paid cloud tiers range from free up to $100 a month, billed by GPU time rather than token count, a structure built around agentic tasks that run in the background.

Has Ollama responded to the enshittification criticism?

Yes. The company maintains that nothing has changed for the free desktop product and frames the paid cloud tier simply as access to models too large for a personal laptop.

Which AI models can I run through Ollama?

The catalog includes open-weight models such as GLM, Nemotron, DeepSeek, Kimi and MiniMax. Partnerships with model labs Meta, Google DeepMind, Mistral and MiniMax give Ollama users access to new releases on their first day out.

Why do regulated industries prefer Ollama’s local option?

Because inference can run entirely on hardware the customer already controls, prompts never have to leave the building. Ollama also says customer data is never used to train its own models, a distinction it markets directly to healthcare, finance and government customers.